Thinking in Images

Also available from Bloomsbury:

Adorning Bodies, by Marilynn Johnson

Aesthetics, Philosophy and Martin Creed, edited by Elisabeth Schellekens and
Davide Dal Sasso
Experimental Philosophy of Identity and the Self, edited by Kevin Tobia
Human Beings and their Images, by Christoph Wulf
The Philosophy and Science of Predictive Processing, edited by Dina Mendonça,
Manuel Curado and Steven S. Gouveia
 Thinking in Images
Imagistic Cognition and Non-propositional Content

Piotr Kozak
 BLOOMSBURY ACADEMIC
Bloomsbury Publishing Plc
50 Bedford Square, London, WC1B 3DP, UK
1385 Broadway, New York, NY 10018, USA
29 Earlsfort Terrace, Dublin 2, Ireland

BLOOMSBURY, BLOOMSBURY ACADEMIC and the Diana logo are trademarks of

Bloomsbury Publishing Plc

First published in Great Britain 2023

Copyright © Piotr Kozak, 2023

Piotr Kozak has asserted his right under the Copyright, Designs and
Patents Act, 1988, to be identified as Author of this work.

For legal purposes the Acknowledgements on p. ix constitute an extension

of this copyright page.

Cover image: Suprematism No. 56, 1916 Kazimir Malevich 1878–1935 Russia,
USSR, Russian, Federation, © Peter Horree / Alamy Stock Photo

All rights reserved. No part of this publication may be reproduced or transmitted

in any form or by any means, electronic or mechanical, including photocopying,
recording, or any information storage or retrieval system, without prior
permission in writing from the publishers.

Bloomsbury Publishing Plc does not have any control over, or responsibility for, any
third-party websites referred to or in this book. All internet addresses given in this
book were correct at the time of going to press. The author and publisher regret any
inconvenience caused if addresses have changed or sites have ceased to
exist, but can accept no responsibility for any such changes.

A catalogue record for this book is available from the British Library.

A catalog record for this book is available from the Library of Congress.

ISBN: HB: 978-1-3502-6746-6

ePDF: 978-1-3502-6747-3
eBook: 978-1-3502-6748-0

Typeset by Jones Ltd, London

To find out more about our authors and books visit www​.bloomsbury​.com and
sign up for our newsletters.
To Kasia, Tadeusz and Józef
vi
 Contents

List of Figures viii

Acknowledgements ix

Introduction 1

1 What is the problem of thinking with images? 17

2 What is thinking? 27
3 What answers should we expect? 55
4 What do images do? 75
5 Recognition-based identification 101
6 What is an image? 123
7 Thinking with images 149
8 Conclusion 181

Notes 185
Literature 196
Index 225
 Figures

2.1 Theories of thinking 50

4.1 The unknot (on the left) and an equivalent knot (on the right) 76
4.2 Reidemeister moves 77
4.3 The left-trefoil knot and the right-trefoil knot 78
4.4 A square pattern matched with a tangram set 80
4.5 The construction of the triangle ∆FGK82
4.6 Line graph representing inflation growth 88
4.7 The pulley system diagram 88
4.8 Line graphs (a) and bar charts (b) convey the same information
but in a different way, affecting the accessibility of the information 92
4.9 Euler diagrams representing the subset, intersection and disjoint
relations, respectively 93
4.10 Two diagrams of a triangle with the same construction rules 97
4.11 Depending on the identified rules of construction, ∆ABD and
∆ACE are similar or congruent to each other 97
5.1 The first picture of a black hole 102
5.2 The cross-ratio invariant 113
5.3 Munker-White’s Illusion 115
5.4 Images impose order on some manifold to localize the elements
of the structure 118
6.1 The 2-D model of iconic reference representing the relationship
between an image, its content, the referent and the target 125
6.2 The relation between trueness, precision and accuracy conditions 134
6.3 Reutersvaard-Penrose triangle 143
6.4 The impossible fork 145
 Acknowledgements

Books never have a single author. They are always the fruit of discussion. I
would like to thank Bartosz Działoszyński, Paweł Gładziejewski, Witold Hensel,
Mateusz Hohol, Małgorzata Koronkiewicz, Paweł Kozak, Mira Marcinów,
Jakub Matyja, Bence Nanay, Kristόf Nyíri, Michał Piekarski, Robert Poczobut,
Robert Rogoziecki and Nastazja Stoch for inspiring discussions and valuable
advice. I am very grateful for the anonymous reviewers for their helpful
remarks. Special thanks to Maja Białek, Krystyna Bielecka, Marcin Miłkowski
and Marek Pokropski, who read the early version of the manuscript for their
insightful comments. I am also grateful to Kamil Lemanek and Krzysztof Gajda
for proofreading.
I wrote this book during the Covid-19 pandemic and finished it when the
Russo-Ukrainian War started. It was not a good time to write books. The only
explanation why this book was written is the continuous support of my family.
Thus, I would like to express my deepest gratitude to the most thoughtful and
caring person I know, who happily is my wife, and the people I love the most,
who happen to be my children.
Chapter 2 includes material from my paper ‘The Diagram Problem’,
Diagrammatic Representation and Inference 2020. LNCS (pp. 217–24). Springer.
Chapter 6 includes parts of my paper ‘The Analog-Digital Distinction Fails
to Explain the Perception-Thought Distinction: An Alternative Account of the
Format of Mental Representation’, Studia Semiotyczne, 35 (1), 73–94.
This work was supported by the research grant ‘What Is Thinking with
Images?’, SONATA 10, granted by the National Science Centre, Poland, based on
the decision No. 2015/19/D/HS1/02426.
x
 Introduction

Let me begin with a philosophical confession of faith on what a philosophical

problem is. The nature of a philosophical problem is ultimately bound up in
knowledge and wonder: granted that knowledge begins with wonder, the nature
of wonder seems to differ between science and philosophy, respectively. In
science, the object of wonder is a discrepancy between theoretical predictions
and experimental findings, where both elements make up a scientific image
of the world. In philosophy, the object of wonder is a discrepancy between
the scientific image of the world and our conceptual intuitions about how the
world looks, which builds the manifest image of the world. The goal of asking a
philosophical question is to overcome this discrepancy, acquiring what Sellars
(2007) calls a ‘synoptic vision’ in which the scientific and manifested image of
the world can be reconciled. The means of achieving this goal – and solving a
philosophical problem – is a conceptual description of the problem that will
allow it to be solved with the help of scientific methods.
As is usually the case with philosophical confessions of faith, the afore­
mentioned credo neither calls for nor provides much in the way of justification.
However, if it is at least partially true – and I believe it is – it helps us to see what
the philosophical problem is with thinking in general and thinking with images
in particular. What makes that last issue philosophically interesting is a deep
discrepancy between the scientific image and the philosophical understanding
of what thinking is and what thinking with images can and cannot be. Examples
of this discrepancy include the problems we encounter when we try to reconcile
two seemingly justified beliefs: that the way engineers, chemists, biologists,
architects or artists use images is essential to their acts of thinking and that the
traditional way we think about thought and thinking, put in terms of inference
relations and the content of propositional attitudes, is non-imagistic in nature.
In other words, we have strong reasons to believe that we think with images and
that this belief cannot be reconciled with the way we think about thinking. I
believe that any attempt to overcome this discrepancy demands reconsidering
our philosophical views on the nature of thinking and the role of images in the
thinking process.
2 Thinking in Images

The main research question of this book is this: What is thinking with images?
The question is analogical to such questions as ‘What is thinking with language?’
It means that if we can ask whether we can think in or with language, then we
can ask whether we can think in or with images.1
The question follows from a commonsensical observation: when we ask how
many windows are in the flat, someone will probably form and inspect a mental
image of the flat and count the windows. If an architect designs a house, then
they design the house with the help of drawings. One may use a map if one tries
to get from point A to point B.
The aforementioned examples are instantiations of what can be called
‘thinking with images’. However, listing examples of imagistic thinking is a
relatively easy task. The difficult task is to say what thinking with images is.
Imagistic thinking is understood here in three ways: as a faculty, an act and a
mental state or an event. The faculty of imagistic thinking refers to the capacity
to use images in thinking. The act of imagistic thinking is exercising this faculty.
The mental state called ‘imagistic thought’ is a product of such an act.
These three understandings of the term ‘thinking’ are interconnected, for
we can only study the faculty of thinking through its expressions in the acts
of thinking. The acts of thinking are interpretable only through studying their
products. Similarly, we can understand what thoughts are only if we understand
what the acts and the faculty of thinking are. Thus, a full-fledged theory of
imagistic thinking should provide a theory concerning the faculty, the act and
the mental state of imagistic thinking.
Imagistic thinking is commonly contrasted with thinking with words (e.g.
Slezak, 2002b; Zhao et al., 2020), for, as is commonly held, not everything
that is depicted or represented in an image can be described in language-like
and propositional form. However, that is only a negative description of the
phenomenon.
The main difficulty stems from the fact that thinking with images is not
a case of trying to solve a clear-cut problem where we need answers to some
ready-made questions. It is the opposite: we need a basic reconsideration and
reorganization of the issue. The fact that we think with images is intuitive but
poorly understood. It is partly caused by the fact that the concept of thinking
with images has always been formulated in various ways, often just in ordinary
language. The problem with spelling out what the expression ‘thinking with
images’ might mean is a crucial part of what thinking with images is. Yet it is
easier to point out what kind of definition we do not want than to produce one
that we would accept.
 Introduction 3

Let me give an example of what such a definition cannot look like. It may
be tempting to hold that iconic representation can be ascribed to the format
of ‘showing’ and the propositional ones the format of ‘saying’. However, it is a
dead end for two reasons. On the one hand, one can overvalue the dissimilarities
and claim that thinking with images expresses the so-called iconic difference,
that is, it is governed by a ‘logic of showing’ which follows different rules (if
any) than propositional logic – a ‘logic of saying’ (e.g. Belting, 2001; Boehm,
2007; Mersch, 2003, 2011; Mitchell, 1994; Müller, 1997). On the other, one may
overestimate the analogies and claim that there is only one format of thinking –
a propositional one (e.g. Fodor, 1975, 1987; Pylyshyn, 2003a). All other formats
are only epiphenomena of the propositional one.
Both approaches are unsatisfactory. The first one is doomed to vague
metaphors and cannot spell out the similarities between propositions and icons.
Images and propositions can indicate a subject and attribute a property to it. If I
take a picture of a man committing a crime, the picture can express a proposition
that can be put into the sentence ‘this man has committed a crime’ (e.g. Kulvicki,
2020; Novitz, 1977).
The second approach is doomed to reductionism and cannot spell out the
differences between propositional and iconic representations. Unlike images,
propositions are governed by the rules of logic. Propositions can be negated and
inferred. Images cannot. Thus, it is essential not to exaggerate the similarities
and not to underestimate the differences.
Notice that the question ‘What is thinking with images?’ is not to ask ‘Do
we think with images?’ One of the assumptions of the book is that we do. The
question is, what does that mean, and what follows from it,
Although the nature and role of (mental) images have been investigated in
many particular branches of philosophy, for example, in logic (e.g. diagrammatic
reasoning), philosophy of mind (e.g. the imagery debate) and aesthetics (e.g.
theory of depiction), what we certainly lack is a general account of what images
are and what role they play in thinking.
The general form of the main question follows from our overall attitude
towards what the philosophical problem is to begin with. As Sellars famously
noted (2007, 369), the aim of philosophy is ‘to understand how things in the
broadest possible sense of the term hang together in the broadest possible
sense of the term’. Thus, although the nature and the role of images are studied
in many specific branches of philosophy, the general aim of philosophy is to
deliver as broad an understanding of the nature and function of images as
possible.
4 Thinking in Images

That is more than just a bold declaration or an expression of faith. Without

an answer to the general question, we would not be able to avoid falling into a
circularity of the following kind: two intuitions form the basis of the very idea
of narrowing down the question ‘What is thinking with images’ to particular
questions, such as ‘What is the epistemological role of images?’, ‘What is the
nature of depiction?’, ‘What are mental images?’ and so on. First, one may claim
that we can indirectly study the general nature of imagistic representations
by studying how we use images within different domains. Second, it might be
said that studying particular kinds of imagistic representations allows us to
put the question of the role of images in thinking in a more precise manner.
However, the claim that we can infer the general features of imagistic thoughts
from studying their particular forms is based on a presupposition that we
can distinguish certain forms of representation as instantiations of some
representational genera. This possibility, however, requires that we possess
at least operational knowledge of what imagistic thoughts are. Still, it is
logically possible that such particular forms of imagistic thoughts instantiate
heterogeneous representational genera. The last hypothesis, however, would be
undecidable if we did not already have knowledge of what the nature of imagistic
thoughts was in the first place. In other words, the point is not that we lack an
answer, for example, to the question of the connection between mathematical
diagrams, scientific graphs, mental images and pictures on a wall, or why we
call them images. The point is that there cannot be an answer based solely on
an analysis of particular phenomena.
The general approach to the question ‘What is thinking with images?’ seems
to be rarely taken, with the notable exceptions of the works of Elgin (2017), van
Fraassen (2008), Goodman (1976), Kulvicki (2014, 2020), Lopes (1996, 2003),
Nanay (2014), Peirce (passim) and Price (1953), to name a few. Such general and
synthetic approaches have always been exposed – most often fairly – to charges
of oversimplification, ambiguity, misuse of metaphors, vague intuitions and
analogies or simple ignorance. Such objections, which experts in narrow fields
of expertise often formulate, often successfully undermine these enterprises.
However, adopting a general approach is essential from the point of view of
the development of science. It provides theoretical space for new hypotheses
and inspires new research directions. Not to mention that it appears that we are
currently stuck with no satisfactory theory of depiction or a universally accepted
theory of the role of images in reasoning. We do not know what thinking is or
how it relates to (mental) images. Thus, it seems reasonable to believe that what
we lack is a synthesis. Some phenomena are visible only from a synoptic point of
 Introduction 5

view. And this book is an attempt to do just that: to look at images and thinking
from above.
There are two contrasting and firm intuitions standing behind the main
question. First of all, there are instantiations of some mental processes, based,
loosely speaking, on using images that we would rightly like to call ‘thoughtful’
but that do not fit into our general view of the nature of thinking. They do not
have a propositional or language-like form, a form expressed with propositions
or sentences.
For instance, we would certainly like to interpret architects’ drawings as part of
their thought process. And similarly, we may want to interpret how engineers,
biologists and chemists use diagrams as an essential part of their scientific
practices (e.g. Rheingold, 1991; Earnshaw and Watson, 1993; Nersessian, 2008;
Tufte, 1997). Scientists think with diagrams (e.g. Ferguson, 1992; Gooding,
1990, 2010; Mößner, 2018; Sheredos et al., 2013; Shin, 2015, 2016; Sloman,
2002). Finally, we would like to say that the result of an artist’s work, such as a
painting or a piece of music, is a result of thoughtful action, even if such results
are put exclusively in pictorial or auditory representational form (e.g. Carvalho,
2019). As Ryle (1979, 85–6) points out:

Mozart’s thinking results in something playable, not stateable. A symphony is

not composed in English or German, it has no translation, there is no evidence
for or against it. It is not grammatical or ungrammatical; neither in prose
nor in verse. Cezanne may make mistakes, but he is not in error. There is no
contradicting the chess player’s carefully or carelessly thought-out move. A
sonnet is not a report, a premise, or a conclusion; it can be a bad sonnet, but it
cannot be a fallacious one.

All the mentioned examples of using images in science and art are hardly
translatable into sentential forms, and it appears that they cannot be put into a
logical form that would determine which step is necessarily next. Therefore, some
acts of thought are hard to reconcile with our general view of what thinking is if
we limit our understanding of thinking to propositional or language-based acts.
Second of all, there is a widespread assumption that images, particularly
visualization, play an essential role in cognition and facilitate solving cognitive
tasks. Most commonly, it is claimed that images are more informative and
basic than sentential representations, epitomized by the maxim ‘a picture is
worth a thousand words’, and that visualization fosters learning and reasoning
processes. However, such generalizations about the benefits of images over
good old-fashioned sentential representations beg the question of what
6 Thinking in Images

is cognitively gained from them. What, for example, is the status of the
commonly held claim that pictures convey more information than any formal
representation or symbol? It is obviously false, for there are symbols, such as
the symbols (1, . . ., ∞), that are infinitely more informative than any picture. It
is not even the case that pictures necessarily convey more visual information.
Some phenomena can be represented symbolically but not pictorially, such as
the concept of an infinite set.
Moreover, it is far from clear what follows from the claim that visualizing
helps in information acquisition (Salis and Frigg, 2020). For instance, how
would visualizing the definition of an integral help us learn how to integrate a
linear function? Visualization is neither necessary nor valuable in that context.
This tension is evident from the perspective of conflicting results in cognitive
science in studies on the effects of visualization in reasoning. Some studies
report that performance on some problem-solving tasks, such as those involving
transitive inference, depends on the ease of imagining the premises of the
reasoning (e.g. Clement and Falmagne, 1986; Shaver et al., 1975). Problem-
solving task participants who build a physical model of a problem are more
likely to solve it. That includes visualizing with the help of graphs and the help
of gestures (e.g. Bocanegra et al., 2019; Goldin-Meadow, 2003; McNeill, 1992;
Vallée-Tourangeau et al., 2016). However, numerous studies have failed to
identify visualization as having any effect on reasoning. For example, Sternberg
(1980) found no difference between the accuracy of solving difficult and easy-
to-imagine problems. There is no correlation between the ability to visualize and
scores on IQ tests and subjects’ ability to reason. Richardson (1987) reported
that individuals reasoning with concrete and easy-to-visualize problems were
no better than individuals reasoning with abstract problems. According to some
scholars (e.g. Byrne and Johnson-Laird, 1989; Knauff, 2013; Ragni and Knauff,
2013), images are irrelevant for reasoning. Human thinking is based not on
abstract symbols or the manipulation of visual images but on spatial cognition,
particularly on multimodal spatial layout models, which are more abstract than
images and more concrete than symbols. It has even been demonstrated (e.g.
Knauff, 2013; Knauff and Johnson-Laird, 2002) that visualization may, in fact,
impede the process of thinking, which is known as the visual impedance effect.
Knauff and Johnson-Laird (2002) show that visual imagery can slow down the
reasoning process. In other words, engaging imagery skills can interfere with
thinking.
The debate, however, is obviously unsolvable if we do not provide a clear-cut
definition of what images and thinking with images are. For example, Knauff
 Introduction 7

(2013) rightly argues that mental imagery may actually hinder reasoning.
However, the claim is only justified if mental images are understood as modality-
specific, visual, conscious, picture-like representations that are supposed to play
the same role as propositions. There is, however, no logical or nomological
necessity for understanding them in this way. In fact, mental images are not
necessary to be visual, conscious, picture-like representations or to play a
proposition-like role.
We encounter the same problem when we assess the value of diagrammatic
representations in the logic and philosophy of mathematics. Although it is rightly
claimed that diagrams play a crucial role in mathematical practice, there is no
consensus on the credibility of reasonings conducted with the help of imagistic
means (e.g. Barwise and Etchemendy, 1996; Giaquinto, 2007, 2008, 2011;
Giardino, 2017; Macbeth, 2012; Mancosu, 2005; Mumma, 2010; Nelsen, 1993).
For example, it is hard to imagine Euclidean geometry without diagrams (e.g.
Macbeth, 2010; Manders, 2008); simultaneously, diagrammatic representations
have been suspected to be unreliable and heterogeneous (e.g. Brown, 1999;
Giaquinto, 2011; Kulpa, 2009).
Moreover, the lack of a general understanding of the nature of images and
their role in thinking may have a negative impact on our understanding of
the role that imagistic representations play within cognitive systems. We risk
making theoretical choices based on tacit knowledge and implicit assumptions
about what images are and what role they can play in thinking without a clear
account of what imagistic representations are. In other words, our theories of
imagistic representations may very well be undercut by conceptual prejudices
rooted in our vague intuitions. For example, it is argued (e.g. Pylyshyn, 2003b)
that what hinders the imagery debate is the lack of agreement concerning what
mental images can be and how we can explicate the phenomenon.
Consequently, we do not know what the empirical findings indicate. Without
a comprehensive account of what ‘mental image’ means, we are condemned to
be immersed in an abyss of vague metaphors and confusing concepts such as
‘functional images’ and ‘quasi-pictures’ that both are and are not images or a
‘mind’s eye’ that is neither in the mind nor the eye. Additionally, it is argued (e.g.
Berman, 2008; Reisberg et al., 2003) that the conflicting intuitions regarding
the nature of mental images may be one of the causes of incompatible data in
the imagery debate. It is plausible that preconceived theories about the nature
of mental imagery influence what subjects say about their experiences, for
theoretical views can and sometimes do influence introspection-based claims
about imagery. Several studies (e.g. Intons-Peterson, 1983; Intons-Peterson and
8 Thinking in Images

White, 1981; Pylyshyn, 2002, 2007) demonstrate that subjects’ performance

in mental imagery tasks heavily depends on the subjects’ expectations and
preconceptions. Experimental results can be significantly distorted by cues
regarding the experimenters’ expectations, additional information and the
subjects’ intuitions (e.g. Hubbard, 1997).
The maxim ‘a picture is worth a thousand words’ does not seem to be
informative regarding the relation between images and language either. It is
commonly assumed that images are essential for using language.2 For instance, in
cognitive psychology, there is a well-established picture superiority effect which
refers to the phenomenon in which images are more likely to be remembered
than words (e.g. Defeyter et al., 2009).
However, without clarifying what we mean by images and words, it is hard to
assess the relationship between them, interpret the available empirical findings
and how these considerations might affect our understanding of language. Let
me give an example of possible confusion this can lead to.
According to the research of Amit and colleagues (2017), the relation between
words and images may be described as asymmetrical. We tend to generate visual
mental images both in the case of visualization and inner speech, but we rarely
think with words when we visualize something. The problem is how to interpret
these findings.
On the one hand, one may say that it follows that manipulating iconic
representations is a more fundamental operation than using language and that
the meanings of the expressions of a given language are built upon the meanings
of iconic representations. On the other, it may mean that imagery is essential in
how we learn the language. For example, we can point at an image of a cat and
express the proposition that is a cat, learning what the word ‘cat’ means. In this
context, this asymmetrical relation between visual and verbal representations
may mark the ontogenetic development of language. It may be a sign of how we
acquire our first language, but it does not imply that it is an essential feature of
the meanings of the expressions of the language in question.
Thus, the value and nature of imagistic representations cannot be assessed
adequately based on our intuitions. Moreover, while there have been numerous
empirical studies investigating different aspects of imagistic representations and
significant theoretical progress on the topic since the end of the last century, few
attempts have been made to integrate the findings into a coherent framework.
As some researchers point out (e.g. Molitor et al., 1989; Scaife and Rogers, 1996;
Winn, 1993), the idiosyncratic character of imagistic representations and their
functions makes it difficult to generalize findings. For example, it is hard to
 Introduction 9

apply the results of studies on the pictorial experience of mental images to motor
mental images, and it is not clear how investigations of the nature of graphs
might apply to pictures. It is difficult to make predictions about the cognitive
value of mental images based on findings concerning the cognitive value of
diagrammatic reasoning or to make predictions about pictures’ cognitive
function based on metaphors’ cognitive function. What is needed is a systematic
approach to interpreting the nature and merits of imagistic representations.
Without such an approach, we have no principled way of making sense of the
vast empirical literature on the cognitive function of (mental) images. Therefore,
what is needed is both an answer to the general question ‘What is thinking with
images?’ and a view that accounts for various uses of imagistic representations
and the interplay between internal (mental) and external representations.

Image definition

Some remarks concerning basic terms are required. When I use the terms ‘image’
and ‘imagistic representation’, I interpret them as synonyms of the Peircean
terms ‘iconic sign’ and ‘iconic representation’.
According to Peirce (CP 1.545-567), representations can be divided into
symbols, icons and indices. Symbols refer to their objects via conventions; icons
share some qualities with their objects; indices are causally interconnected with
the represented objects. This trichotomy is based on the form of the relation
between the representation and its object. The distinction denotes different ways
in which this relation can be founded, referred to by Peirce as the ‘ground’ of the
sign (Short, 2007). We can speak of symbols, icons and indices, depending on
the identified ground of the relation – an arbitrary convention, a similarity or a
causal interconnection.
However, a decent interpretation of Peirce’s theory of icons shows that it is
disputable whether Peirce holds that the concept of resemblance is sufficient to
explain how icons represent (Ambrosio, 2014; Chevalier, 2015; Hookway, 2000,
2007; Pietarinen, 2006, 2014; Stjernfelt, 2000). For the reasons mentioned later,
it is better to understand the notion of ‘image’ as a representation that bears a
natural and direct relation to the represented object (e.g. Burge, 2018; Giardino
and Greenberg, 2015). This description is motivated by the observation that
there is, without specifying it now, a natural relation between images and
depicted objects which is believed to be based on exemplifying certain features
of said objects. In comparison to the language, it is not accidental that a dog’s
10 Thinking in Images

picture is like a dog, where it is largely irrelevant whether we call a dog ‘dog’ or
‘Hund’. The words ‘dog’ and ‘Hund’ are not dog-like; the image of a dog is. In
comparison to natural signs, such as animal tracks in the snow, images do not
have to be causally mutually connected with the represented object. An image
of Santa Claus does not have to be caused by Santa Claus, while animal tracks
have to be caused by animals. If there were no animals, there would be no animal
tracks. We cannot say the same about Santa Claus.
Importantly, we cannot determine the sign’s ground outside of the context
of how the sign is taken. The letters p and b can be considered icons if we
pick out the spatial similarities between them. They can be considered different
symbols if we isolate the conventional relation by which they represent. Thus,
depending on the interpretation of the relation between representations and
represented objects, we can take these letters as instantiations of icons or
symbols. Consider the diagram of a triangle ∆. Depending on the way it is
interpreted, it can be an icon; a symbol, for example, ∆ABC; or an index, for
example, a symptom of the mental illness manifested by compulsive drawing
of triangles.
Following the Peircean tradition, I do not restrict the definition of images
to visual representations. One can talk about tactile (e.g. Lopes, 1997; Kulvicki,
2006b; Yoo et al., 2003), haptic (e.g. Klatzky et al., 1991), olfactory (e.g. Bensafi
et al., 2003, 2005; Djordjevic et al., 2004, 2005; Gilbert et al., 1998), gustatory
(e.g. Croijmans et al., 2019; Kobayashi et al., 2004) and auditory images (e.g.
Halpern et al., 2004; Hubbard, 2010; Jakubowski, 2020; Zatorre et al., 2010).
They are just as common and just as psychologically important as visual ones
(e.g. Sebeok, 1979; Newton, 1982). When addressing questions like ‘How do
you clean a window?’ or ‘How would you feel if you won a Nobel Prize?’, one
can form a motor (e.g. Guillot, 2020) and an emotional image (Blackwell, 2020;
Nail, 2019; Thagard, 2005). We can use mental images to think about temporal
relations (e.g. Viera and Nanay, 2020). Maps are imagistic representations of
spatial relations. And diagrams are images spatially representing non-spatial
relations. In a broad sense, even language is describable in pictorial terms. For
instance, visual metaphors or graphic descriptions may be interpreted as literary
devices that form a literary image or contribute to a literary image (e.g. Collins,
1991; Scarry, 1999; Troscianko, 2013).
Such a description is far from clear, and it comes as no surprise that there is
no accepted definition of an image among psychologists and philosophers. That
being said, an explication of the concept of an image is one of the aims of this
book.
 Introduction 11

We can begin by introducing two constraints to help us produce a positive

account. Based on this broad understanding of the term ‘image’, we can see, on
the one hand, that any explanation of the imagistic genera should go beyond the
experiential properties of images; on the other, it should be able to capture the
role of the phenomenological properties of their content.
According to experiential accounts of depiction (Briscoe, 2016; Budd, 1995;
Gombrich, 1960; Hopkins, 1998; Newall, 2011; Peacocke, 1987; Podro, 1998;
Voltolini, 2015; Wollheim, 1987, 1998), perceiving objects displayed with
images elicit a perceptual experience phenomenologically similar to seeing
the represented objects in the flesh. However, experiential accounts are too
restrictive to capture the concept of an image.
Let me illustrate it with Gregory’s (2013) theory of distinctively sensory
content. According to Gregory, the distinctively sensory content of images is
subjectively informative. It identifies what it is like to be in a certain sensory
state. It is perspectival; it specifies a perspective from which the sensory content
of an image can be mapped onto the sensory content of an appearance.
The advantage of Gregory’s definition is that it is broad enough to capture
non-visual modalities. However, the definition of the content of images based on
subjective informativeness and perspectivity is too narrow. First, these conditions
seem to be insufficient. Suppose that a black canvas causes the common illusory
experience of seeing Marilyn Monroe (Schier, 1986, 197). Still, it does not depict
her. Second, we can speak of unconscious images deprived of phenomenological
content. Traditionally, we define mental imagery in experiential terms (e.g.
Finke, 1989; Thomas, 2003, 2009), but it is unnecessary to do so. It is logically
and empirically possible to have a mental image we are not conscious of. It is
logically possible since if there is no contradiction in the concept of soundless
speech, then there is no contradiction in the concept of an unconscious image. It
is empirically possible as there are subjects who are believed to have no conscious
mental imagery experience and are still capable of performing mental imagery
tasks (e.g. Zeman et al., 2010).
Moreover, if imagistic thoughts were only conscious experiences, we would be
unable to capture unconscious mental events. Given that there are unconscious
thoughts, it is implausible that all imagistic thoughts have to be experienced.
And even if that were the case, the burden of proof would be on the proponents
of the view that images are always conscious.3
With that in mind, any theory of thinking with images must assume that it is
possible that there are unconscious images. Therefore, experiential properties of
imagistic content cannot be a criterion for defining images.
12 Thinking in Images

However, abstracting away from the phenomenology of images is a dead end

too. For instance, we can think of images in terms of the structural features of
the format of representation (Dretske, 1981; Goodman, 1976; Haugeland, 1998;
Kulvicki, 2006a) or in functional terms as props in so-called games of make-
believe (Walton, 1990). Yet any theory of depiction should be able to explain the
role of the experiential properties of images. If one were unable to do so, it would
be considered insufficient.
Following the terminological practice of the theory of depiction (e.g. Kulvicki,
2014), I distinguish the term ‘image’ from ‘picture’. ‘Image’ is a general term
referring to all iconic representations. Pictures are a proper subset of images.
When I use the term ‘picture’, I refer to external representations instantiated
by photographs, paintings and drawings. Images comprise external (pictures)
and internal representations (mental images) that are postulated to have iconic
properties. The difference between internal and external representations
lies in the nature of the vehicle of representation: either the representation is
constructed in the mind, or it is based on an external and public medium, such
as paper (e.g. Shin, 2002; Stenning and Yule, 1997).4

Imagistic thinking

The main problem with describing what thinking with images is stems from
the fact that the general way in which images are described, including visual,
auditory, olfactory, literary images, and so on, makes it difficult to read off
the properties of imagistic thinking directly from the properties of images. In
contrast, we can read off the properties of thoughts from the logical properties
of language.
It is easier, however, to say what thinking with images is not. Imagistic
thinking is contrasted here with propositional and language-like thinking. The
term ‘propositional thinking’ refers to how we think about thinking and how
we interpret the nature of thoughts. Thoughts are interpreted as contents of
propositions. The thought that snow is white is the content of the proposition
snow is white. Thinking is, first, the way we stand in relation to certain
propositions, such as snow is white, which is most often interpreted as holding
a propositional attitude, such as ‘I believe that snow is white’ or ‘I doubt that
snow is white’ (e.g. Bermúdez, 2003; Peacocke, 1986),5 and, second, as the way
thoughts stand in relation to each other, which helps us to explain the nature of
logical inferences between thoughts (e.g. Braine and O’Brien, 1998).
 Introduction 13

Propositional thinking is closely bound to language-like thinking. It is

believed (e.g. Bermúdez, 2003; Devitt, 2006; Frege, 1984; Rey, 1995; Sellars, 1969)
that for any propositional attitude, there should potentially, yet not necessarily
actually, exist a sentence that makes the content of the attitude expressible. It
means that propositional thoughts have linguistically expressible contents. It
follows from the observation that thoughts are not directly interpretable. A
thought has to be represented in an interpretable medium to grasp it. Language
is the only medium that can mirror the logical structure of thought and express
the richness of thought’s content. Therefore, propositional thinking requires
language.
In contrast, imagistic thinking is non-propositional, which means that it
is not a matter of standing in some relation to a proposition and has content
that is expressible in images. Imagistic thinking is irreducible to propositional
thinking, just as it is believed (e.g. Ben-Yami, 1997; Crane, 2001; Grzankowski,
2013; Merricks, 2009; Montague, 2007) non-propositional states are irreducible
to propositional ones. Thus, thinking with images is understood here as a kind
of thinking that is non-propositional and expressible in images.

Thinking with Images (Imagistic Thinking): the kind of non-propositional

thinking, the content of which is expressible in images.

Paradigmatic examples of thinking with images are the way musicians use
auditory images to think about music; the way painters use pictures to express
thoughts about perceptual qualities; the way architects employ drawings to
reason about spatial relations; the way scientists and engineers use diagrams and
sketches in scientific reasoning; and the way people employ imagery skills.
According to the imagistic theory of thought, some thought processes are
necessarily based on rule-governed mental or extramental image sequences.
Images are the building blocks of some thoughts, and they determine the content
of those thoughts.

A measurement-theoretic account of thinking with images

I will argue that the imagistic theory of thinking can be defended within
the framework of measurement theory. The basic idea is that we can apply
measurement-theoretic concepts to the analysis of imagistic thinking in the
same way as we apply them to study propositional attitudes (see Dennett, 1987,
1991b; Dresner, 2010; Marcus, 1990; Matthews, 2007, 2011).
14 Thinking in Images

In a nutshell, I claim that thinking is an operation that locates an item in

some logical or physical space, that is, an operation that enables us to localize
and orient the item in relation to some parameters, such as volume, pressure,
temperature, colour and shape, with the help of which we can recognize the
item. Thinking is a manipulation of those parameters, or ‘turning the knobs’, as
Dennett (2013) puts it. It is, for example, a matter of asking about what would
change if some parameters were different, localizing the object of inquiry in a
new measurement-theoretic set-up.
At the same time, thinking is not just any kind of operation. In contrast to
dreaming and hallucinating, it is an operation that can be correct or incorrect (see
Sellars, 1949). Consequently, imagistic thinking is a rule-governed operation.
An analysis of imagistic thinking is an analysis of these rules.
According to the measurement-theoretic account of images, images can
be interpreted in terms of being measurement devices, comparable to rulers
and balances. Images exemplify constructions by means we localize objects.
Consequently, I argue thinking with images is a rule-governed manipulation of
construction rules revealing how something can be perceived in a measurement
set-up, for instance, thinking about how an object would look if a parameter,
such as height or length, were different. In short, thinking with images is a skill
in using construction rules, comparable to using rulers. Thinking with images
reveals the ways the world can be perceived and measured.
This sketch obviously requires further elaboration. It points us in the direction
we are heading rather than presenting the final picture. The rest of the book is
devoted to explicating just that.
However, one point raises no doubts. Studying the nature of images can
teach us something important about thinking, in the same way as studying the
nature of language reveals part of the nature of thinking. Thus, it is plausible that
investigating the nature of images may play the same role in the philosophy of
mind as investigating the nature of language. Suppose it is true, as it is claimed
by most analytical philosophers from Frege to Dummett, that understanding the
nature of language can bring us closer to understanding the nature of thoughts.
Then, plausibly, the same can be said about understanding images.
The only difference is that although there is a well-established tradition of
studying thought processes through studying the nature of language, similar
analyses in the case of images are relatively underdeveloped. In other words, we
know much more about language than about images and, consequently, about
thinking with language than about thinking with images. This book takes a step
towards changing that.
 Introduction 15

Content overview

Dennett has once said that philosophers are better at asking questions than
answering them. I am not convinced that this is entirely true. However, I am
certain that a clear formulation of a problem is a crucial part of a philosopher’s
toolkit.
Following this line, the book consists of two parts. In the first part
(Chapters 1–3), I explain the nature of the problem of thinking with images.
In the first chapter, I present the problems related to the idea of thinking with
images. I argue that explaining the nature of thinking with images requires
proving two theses. According to the Irreducibility Thesis, images are necessary
for (some) thinking processes. This involves showing that images can be bearers
of thoughts and express a non-propositional kind of content and knowledge.
According to the Translatability Thesis, some imagistic thoughts and contents
can be expressed by propositions. I hold that both theses can be explained and
reconciled by adopting the measurement-theoretic account of images, according
to which images are measurement devices, just like rulers and balances.
In the second chapter, I show how thinking about the nature of thoughts
and images can lead to conceptual problems of thinking with images. Images
cannot provide any theory of knowledge and content. They cannot explain the
systematic and decomposable nature of thoughts. I introduce the so-called
Received View, which is a set of beliefs and intuitions regarding the nature of
thinking and which can be framed in a phrase that thinking is propositional
and action-based. I demonstrate that the Received View is incompatible with
the so-called Traditional View regarding the nature of images, which holds that
images resemble represented objects.
In the third chapter, I explore the kind of answers we can expect by asking
questions about the nature of thinking with images. I present two strategies for
answering these questions. According to the neo-Lockean approaches, imagistic
thoughts are abstract entities that mediate between perceptual and discursive
representations. According to Wittgenstein–Ryle’s line of argument, the question
about the imagistic vehicle we think with is senseless. I show that both strategies
fail. I introduce the so-called operational approach, according to which we can
think of imagistic thoughts as the kinds of operations that require images to be
expressed. The question about the nature of imagistic thoughts is the question of
the nature of these operations.
In the second part (Chapters 4–7), I present a positive proposal for addressing
the problem of thinking with images. I do this by analysing two case studies. In
16 Thinking in Images

the fourth chapter, I study the content of knot diagrams. To explain how they
represent, I introduce the concept of construction, which refers to the procedures
of arriving at a target by determining the parameters of some logical or physical
space. I show that the concept of construction is crucial for our understanding
of how images represent.
In the fifth chapter, I analyse the black hole picture. In order to explain its
semantics, I introduce the concept of recognition-based identification. I hold
that recognition is distinct from having beliefs. It is a kind of reference that is
also different from demonstratives and descriptions. Recognition consists in
identifying construction invariants.
In the sixth chapter, based on the concepts of construction and recognition,
I present the so-called two-dimensional (2-D) model of iconic reference.
According to this model, images denote their targets and exemplify the rules of
construction that identify the referent. I show that the best way to explain the
semantical properties of images is to think of them as measurement devices.
In the seventh chapter, I show how the measurement-theoretic account of
images can address the challenges described in the second chapter. I argue that
the 2-D model provides us with a theory of knowledge and content. It allows
us to explain the systematic and decomposable nature of thoughts. Last but not
least, I show the metaphysical consequences the measurement-theoretic account
of images has on our thinking about the nature of the mind. I apply the 2-D
model to explain the nature of the representational format and mental imagery.
This book is long. Although I highly recommend reading it in the order
presented in the consecutive chapters, it can be read in at least two ways depending
on the subject of interest. Those who are intrigued by the philosophy of mind
can go to Chapters 2–3 and 7. Those interested in the theory of depiction can
focus on Chapters 4–6, where I present a model of iconic reference. However,
this book should not be taken as presenting a full-fledged theory of depiction.
That would require giving a detailed context of a discussion, explicating the
positions I am arguing against and reasons why my model is better than others.
The reader will not find it here since it would make the book unreadable. My goal
is more moderate. I take the expression that we think with images for granted
and investigate its consequences for our conceptions of thinking. I believe that
the best way to make this expression sensical is by accepting the measurement-
theoretic model of images.
 1

What is the problem of thinking with images?

Trying to answer the question ‘What is thinking with images?’, we have to

address two philosophical problems: the problem of the alleged instrumental
role of images and the problem of the epiphenomenal nature of images. The first
problem can be explained as follows: thinking with (mental) images is one of the
recognized forms of non-verbal thought in contemporary cognitive psychology
(Gardner, 2004; Kosslyn et al., 2006; Paivio, 1963; Shepard and Metzler, 1971)
and neuropsychology (Greenberg et al., 2005), is basic for scientific reasoning
(Abrahamsen and Bechtel, 2015; Bechtel, 2017; de Regt, 2014; Gendler, 2004;
Meynell, 2018, 2020; Swoyer, 1991; Thagard, 2012),1 is common in engineering
(Ferguson, 1992) and in mathematics (Giaquinto, 2007), is indispensable for art
(Arnheim, 1969) and is essential for common practical reasoning such as ‘If I do
[image], then I shall be able to do [image]’ (Gombrich, 1990).
However, a time-honoured view is that the utility of images is only
instrumental and cognitively inferior. Knowledge, traditionally conceived,
consists of true and justified propositions. Knowledge is most commonly
expressible in language, for only language can express logical relations. Images
have limited expressive power; they are unreliable and cannot be trusted for
inferences. In sum, they do not fit into a propositional account of knowledge.
Images can only illustrate ideas or be used for the purposes of communication.
Images can have a facilitating role and some psychological impact. But that is all.
They do not have any genuine epistemological value.
One of the profound challenges in contemporary epistemology is to prove
that images provide an irreplaceable and autonomous form of knowledge. Such
a demonstration must not be confused with the dominant view that visualization
is an inferior and illustrative presentation of cognitive processes and arguments,
which have, for their own part, a source that is entirely independent of such
visualization (e.g. Botterill and Carruthers, 1999).
18 Thinking in Images

The second philosophical problem with thinking with images concerns

the alleged epiphenomenal nature of imagistic thoughts. Both problems are
interconnected. First, if we accept the claim that the cognitive role of images
is instrumental, then it follows that images may be epiphenomenal. They can
accompany our thoughts while remaining inessential to the thoughts themselves
and are ultimately unable to affect the contents of thoughts.
Second, if we agree that imagistic thoughts have an epiphenomenal nature,
then it follows that they can only have an instrumental cognitive role. They may
be widely used in cognition, much like pencils are widely used in engineering,
but they are unnecessary for cognition, just as pencils are unnecessary for
engineering. Images may be useful in cognition, just as pencils may be useful,
but they do not ground cognition. They may prompt some thoughts, but they
cannot determine the content of thoughts.
Historically speaking, there is at least one reason to believe that images
play a crucial role in thinking. The so-called image theory of meaning and
cognition was prevalent in modern philosophy and early cognitive psychology.
It was believed that images could give us an answer to the question of concept
acquisition. It was also widely believed that images possess iconic content, which
characterizes perception.
However, both twentieth-century analytic philosophy and psychology are
based on the rejection of images. Frege opens his seminal paper Thoughts (1984)
with the constatation that images cannot be vehicles of thought. The roots of
modern psychology can be traced back to the so-called imageless thought
controversy and the birth of behaviourism. In Psychology as the Behaviorist Views
It, Watson’s (1913) example of imageless thoughts is given as among the first
in justifying the critique of Wundt’s introspection-based school of psychology.
Due to twentieth-century critiques (e.g. Bennett and Hacker, 2003; Pylyshyn,
2003a; Ryle, 1949; Wittgenstein, 1953), it is now held that images cannot be
representationally basic, and the phenomenon of thinking with images has come
to be regarded as epiphenomenal and peripheral in cognition. The critique stems
from two partially independent sources.
For one thing, it has been argued that images cannot determine the content
of thought and that the central processes of thought require a propositional
representation system. Images are considered epiphenomena of some abstract
processes of thinking, conducted, for example, in some language-like symbol
system.
For another, the view that thought might consist of disembodied images and
that the mind is a sort of a private motion-picture theatre has always seemed
 What is the Problem of Thinking with Images? 19

philosophically dubious (e.g. Dennett, 1981a, 1991a; Goodman, 1984, 1990).

Mental images can be studied by science, but as such they do not belong to the
scientific image of the world. They are a part of its manifest image (Dennett,
1981b, 2017). In consequence, as was explicitly stated by Candlish (2001, 108):

Now, however, it has become so thoroughly absorbed that it has disappeared from
philosophy’s surface and an imagistic account of thinking such as is outlined in
Russell’s Analysis of Mind (Lecture X) or elaborated in H. H. Price’s Thinking
and Experience is usually no more felt to deserve critical attention than is, say, a
geocentric account of the universe.

Let me underline this point. The imagistic theory of thought has been replaced
with the propositional one not because a competing theory with a more
explanatory power has been introduced. The propositional theory of thought is,
in many aspects, no better than the imagistic one. Instead, the imagistic theory of
thought appeared to be internally inconsistent, and it was accordingly rejected.
As Sellars (1963, 15) puts it: ‘all attempts to construe thoughts as complex
patterns of images have failed, and, as we know, were bound to fail.’
However, regardless of the staunch rejection of the imagistic theory of
thought, even its harshest critics agree that even if not all thought processes are
conducted in images, some of them are or at least that some thought processes
do involve (mental) images. For instance, it is believed that images are crucial
to explaining the nature of animal thinking (e.g. Burge, 2010; Gauker, 2011;
Mullarkey, 2011) and the nature of concept formation (e.g. Barsalou, 1999;
Carey, 2009). Whatever may be meant by the claim that mental images do or do
not exist, mental imagery is a real phenomenon and quite open to quantitative
scientific investigation (e.g. Shepard, 1990). It is claimed (e.g. Block, 1983a;
Gauker, 2011; Kaufmann, 1996; Rollins, 1989) that the debate on the possibility
of thinking with images challenges the dominant view of the nature of thinking.
However, we still lack a clear understanding of what we mean by the imagistic
theory of thought (e.g. Abel, 2012; Bechtel, 2008).

The Irreducibility Thesis

The imagistic theory of thought can be explicated in terms of the Irreducibility

Thesis, which holds that images are necessary in thinking. It consists of the
conjunction of the Metaphysical, Semantical and Epistemological Irreducibility
Theses:
20 Thinking in Images

The Irreducibility Thesis: images are necessary in thinking.

The Metaphysical Thesis: Images are bearers of thoughts;

The Semantical Thesis: Images possess a kind of content distinct from that of
non-imagistic representations;

The Epistemological Thesis: Images are vehicles of a kind of knowledge that

cannot be obtained via non-imagistic means.

The Irreducibility Thesis expresses the belief that imagistic representations

are neither epiphenomenal nor peripheral in the process of thinking. It
means, first, that the existence of some thoughts depends on the existence
of imagistic representations expressing these thoughts. Second, images and
imagistic thoughts possess iconic content that is not reducible to other forms
of content, in particular to propositional content. There is uniquely imagistic
non-propositional content. Third, images and imagistic thinking do not provide
merely instrumental and inferior forms of knowledge. Some cognitive functions
are available only in virtue of the use of imagistic representations. It is not only
that we can learn something from images, but images are a source of knowledge
that is not available via non-imagistic means. The aim of the book is to show that
the Irreducibility Thesis is true.
To illustrate the Irreducibility Thesis, think about how people keep
photographs of their kids in their wallets. For some reason, we keep photographs
of our kids and not name cards or their extended descriptions, although kids
can be described with the help of propositions such as he or she is blond,
and so on. We want to say that the photograph represents something more. The
question is what that something is.
One remark concerning the scope of the Irreducibility Thesis is needed. I argue
that there are instantiations of the irreplaceable role of images, but I do not claim
that every use of images is of that kind. There are certainly merely illustrative and
instrumental uses of images. However, the Irreducibility Thesis holds that there
are uses of images in accord with the thesis. Thus, even if images can sometimes
be replaced in thinking, that does not mean that they are always replaceable.

Two interpretations of the Irreducibility Thesis

There are two ways of describing the irreducible role of images. First, we can
understand the role of images weakly in terms of the facilitating role of images.
 What is the Problem of Thinking with Images? 21

That means that the basic role of images is to prompt some thoughts and facilitate
intellectual processes. Images can be interpreted as a necessary condition of
some thought processes, where necessity is understood in epistemological terms:
images are necessary to grasp some thoughts and acquire information. In the
same way, computers are necessary to conduct some complicated calculations.
Second, we can understand the role of images in strong terms, where images
are interpreted as constitutive elements of cognitive systems. According to the
Strong Interpretation, the existence of some thoughts is necessarily bound with
the existence of images. Necessity is understood in metaphysical terms, which
means that it is impossible to have some thoughts without an image. As an
example, numbers are necessary for counting; without numbers there would be
no elements we could count with.
We can distinguish between two interpretations of the imagistic theory of
thought with respect to each of the above. According to the Weak Interpretation,
some cognitive processes, such as learning, reasoning or problem-solving, are
based on iconic representations. For example, it is held that the use of mental
images is crucial to solving mental rotation tasks or that there is a well-
established relationship between the ability to use diagrams and success in
mathematical problem-solving tasks (e.g. Casati, 2018; Grandin, 2006; Hegarty
and Kozhevnikov, 1999; Thagard, 2005; Tversky, 2011, 2015).

The Weak Interpretation of the Imagistic Theory of Thought (Weak Interpretation

for short): Images are epistemologically necessary for grasping the content of
some thoughts.

That being said, the Weak Interpretation is insufficient to explicate the imagistic
theory of thought. Let me demonstrate that using Larkin and Simon’s theory of
diagrammatic reasoning.2
According to Larkin and Simon (1987), diagrams can be distinguished
from sentential representations by the way they display information. Diagrams
provide a one-to-one mapping of information stored in a spatial form at the
particular locus of a diagram, including information about relations with the
adjacent loci. We can infer the features of the represented objects by inspection
of the spatial features of the representational vehicle.
In comparison to sentential representations, diagrams are cognitively less-
loaded. They represent more pieces of information at once. They help in solving
deductive and abductive tasks (e.g. Coopmans, 2014; Kirsh, 2010; Kitcher and
Varzi, 2000; Latour, 1990; Zhang, 1997), for the process of acquiring information
is perceptually enhanced by the way the information is organized (e.g. Bauer
22 Thinking in Images

and Johnson-Laird, 1993; Beilock and Goldin-Meadow, 2010; Bordwell, 2008;

Giardino, 2014, 2016; Ware, 2000).
Consequently, diagrams are the source of a ‘diagrammatic’ form of inference
(e.g. Barwise and Shimojiima, 1995; Gurr et al., 1998; Lemon et al., 1999;
Lindsay, 1998; Shin, 1994, 2002; Stenning, 2002; Stenning and Lemon, 2001;
Stjernfelt, 2007). While symbolic reasoning is based on the interpretation of
abstract symbols, a diagrammatic form of inference is based on the fact that we
see the conclusion outright. What would be an active inference from premises
to a conclusion in a symbolic representation system comes along as a ‘free ride’
in diagrammatic systems (e.g. Larkin and Simon, 1987; Shimojima, 1996, 2015).
According to Larkin and Simon, the difference between diagrams and
sentential representations is a matter of different notation. Diagrams use spatial
representations to express logical relations. Sentential representations are based
on symbols. Yet, in principle, both representational systems can express the
same information; they are informatively equivalent.
The difference in notation is not trivial. Some notations are more efficient
than others, and much of the progress in mathematics is based on discovering
different notational systems. Additionally, if the difference between diagrams
and sentential representations is a matter of notation, it is relatively easy to
prove that diagrams may be more efficient than sentential representations.
The difference in the efficiency of reasoning can be interpreted in terms of the
number of computational operations that need to be performed to move from
the initial state to the goal state. Larkin and Simon’s proof of the computational
efficiency of diagrams does exactly that.
Although Larkin and Simon’s theory succeeds in describing the epistemological
merits of images, it is insufficient to explicate the imagistic theory of thought.
Note that Larkin and Simon do not claim that diagrams are irreplaceable
in thinking. Diagrams are good instruments for thinking with. They may be
functionally necessary for some inferences, in the same way as computers are
functionally necessary to calculate large numbers. Yet the same information
that is encoded in diagrams can be expressed with symbolic representations.
Sentential and diagrammatic representations are equivalent in the information
they provide. They are not computationally equivalent.
The main problem with the instrumental view Larkin and Simon provide is
that it does not exclude that images are parasitic on other forms of representations.
Images can accompany the act of thought, they can help us grasping the content
of thought, but they are not part of the logical mechanisms of thought (Kozak,
2020). Images, in accordance with Larkin and Simon’s theory of diagrammatic
 What is the Problem of Thinking with Images? 23

reasoning, can efficiently make the content explicit. Images can draw attention
to unnoticed properties of the content. They can prompt different ways of
perceiving objects. Yet, they do not determine the content of thoughts. They can
make some thoughts more precise and point out some features of the content, but
they cannot change the content. In the same way, a microscope is a valuable tool
to make evident the content of a sample, but it does not determine the sample’s
content. A microscope does not create microbes, but it may be necessary to
reveal them. Larkin and Simon’s argument demonstrates that images can play a
crucial role in grasping the content of some thoughts but does not prove that we
think with images.
Let us turn to the Strong Interpretation. According to the Strong Interpretation,
imagistic thoughts can be distinguished by the nature of the operation. The
nature of imagistic thought is interpreted in terms of a family of rule-governed
processes. To ask about the nature of the operation is to ask what the operation
does. For example, if one is interested in the nature of a quadratic function, then
one is interested in how one operates with some variables. If one understands
what a quadratic function is, one understands what it does. In the same way, if
we are interested in the nature of imagistic thoughts, we are interested in what
they do. In a nutshell, some operations could not exist if there were no images,
in the same way as some mathematical operations would not exist if there were
no mathematical objects. The question is what these operations are.
Accordingly, we can define the Strong Interpretation of the imagistic theory
of thought.

The Strong Interpretation of the Imagistic Theory of Thought (Strong Interpretation

for short): Images are metaphysically necessary for the existence of some
thoughts.

The Strong Interpretation holds that images are irreplaceable in thinking. The
irreplaceability of images means that some thoughts do not exist, if there are
no images expressing them. In other words, some thinking processes can be
identified with genuinely imagistic operations. The main task here will be to
establish this irreplaceable role of images.

The Translatability Thesis

One of the main problems with the Irreducibility Thesis is that it has to be
reconciled with the Translatability Thesis. According to the Translatability
24 Thinking in Images

Thesis, imagistic thoughts can be translated into propositional ones. It means

that some imagistic content can be expressed with the help of propositions. The
idea is that we can easily go from images to words and the other way round.
Words can be used to describe images. Images can be used to make statements,
warn or criticize (e.g. Eaton, 1980; Kjørup, 1974; Kulvicki, 2020). Any theory of
imagistic thinking has to take it into account.
The Translatability Thesis consists of the conjunction of two claims. The
Semantical Translatability Thesis expresses the intuition that any content of
an image can be expressed with a potentially infinite sequence of propositions
describing the represented object. For instance, every map can be translated into
a set of propositions in the form of verbal coordinates expressing the location
and relations between spatial objects. The content of a picture can be described
verbally, for example, with the proposition that it represents a woman with blond
hair. Sounds can be expressed in a propositional form represented with notes,
such as sound A follows B and so on.
The Epistemological Translatability Thesis states that knowledge expressed with
images, such as ‘if I do so [image], then I will be able to do so [image]’, is expressible
in propositional form, such as ‘if I move the pedals clockwise, I will be able to ride
a bike’. Knowledge of what my aunt looks like expressed with the help of a memory
image can be translated into a description that my aunt is blond and so on.
The Translatability Thesis: Some imagistic thoughts can be translated into
propositional ones.

The Semantical Translatability Thesis: Some imagistic information can be

expressed by propositional representations.
The Epistemological Translatability Thesis: Some imagistic knowledge can be
expressed by propositional knowledge.

The reason for accepting the Translatability Thesis is that we want to maintain
correctness conditions for imagistic thinking, which means that we can be
wrong when thinking imagistically (e.g. Dilworth, 2008; Langland-Hassan, 2015,
2020). First, we want to maintain the idea that it is possible to make a mistake
in describing the content of an image. For example, it would be a mistake to
say that a photograph represents a blond girl if it represents a dark-haired man.
If propositions and images were untranslatable, then it would be impossible to
say that they fit each other, in the same way as it would be nonsensical to ask
whether a random set of letters describes a girl.
Second, we want to maintain the idea that imagistic representation may
correctly or incorrectly represent the world. If I depict a wanted thief as a blond
 What is the Problem of Thinking with Images? 25

girl and the thief turns out to be a dark-haired man, I can say that I was wrong
because the proposition the thief is a dark-haired man is true. If imagistic
and propositional thoughts were untranslatable, it would be difficult to say in
what sense I could be wrong when I depict the thief as a blond girl.
Most theories of imagistic thoughts are in tension either with the Irreducibility
Thesis or the Translatability Thesis. There is a tendency to swing between the
two. On the one hand, imagistic thoughts are sometimes believed to be a kind
of propositional form of representation (e.g. Pylyshyn, 2003a), which makes it
unclear how to express the genuine value of imagistic representation. On the
other, imagistic thoughts are believed to be non-propositional (e.g. Gauker, 2011;
Mößner, 2018), but it is unclear how we can translate the content of imagistic
thoughts into propositional content. Hybrid views are often formulated (e.g.
Camp, 2007, 2015, 2018; Denis, 1991; Fodor, 1975; Kulvicki, 2020; Langland-
Hassan, 2015), which state that a proper part of the content of imagistic
representation is propositional. However, it is not easy to see what, according to
hybrid views, the contribution of the iconic content of thoughts is to cognition
and what functions it plays in the broader cognitive economy.
Let me illustrate the problem with hybrid views using Fodor’s theory of mental
images. Fodor (1975, 2008) holds that images have to be put under description
to fix their meaning which then allows them to be implemented in the mental
mechanism. A sentence in the language of thought (LOT) has to be attributed to
imagistic content. In a nutshell, if we introduce a system of mental symbols, such
as LOT, and assign these symbols to images, we can determine the meaning of
images and incorporate images into the machinery of thought.
However, Fodor’s hybrid strategy renders imagistic representations redundant.
If a discursive representation determines the content of images, then in principle,
imagistic representations have no genuine role in cognitive architecture and can
be reduced to propositions. That is, if we assign each image a sentence in LOT
and define thinking operations as operations on LOT’s symbols, then images do
nothing in the mind’s machinery. Images can play a role in grasping the content
of some thoughts, but they are not necessary for having any thoughts.
In this book, I argue that any attempt to defend a non-trivial imagistic
theory of thought has to take into account both the Irreducibility Thesis and the
Translatability Thesis. It means that imagistic thoughts have to be expressible
in propositions and have to carry such information that propositions cannot
express. Most of this book will be devoted to showing how the Irreducibility and
Translatability Theses can be comprehensively explicated in the context of the
imagistic theory of thought. I shall argue that the former can be defended based
26 Thinking in Images

on a measurement-theoretic account of imagistic representations. In a nutshell,

we can think about images in terms of measurement devices. Therefore, we can
defend the genuine role of images in thinking without being committed to the
view that thinking consists in forming propositional representations of imagistic
content.

Summary

Thinking with images is the kind of non-propositional thinking the content of

which is expressible in images. The problem of thinking with images is how to
reconcile a set of contradictory claims: on the one hand, it is claimed that images
play an essential part in cognition and, on the other, that they are instrumental
and peripheral in thinking.
According to the imagistic theory of thought, the existence of images is
necessary for the existence of some thoughts. The imagistic theory of thought
may be explicated in terms of the Irreducibility Thesis, according to which images
are bearers of thought, possess a distinct kind of content and provide knowledge
that cannot be obtained via non-imagistic means. The main problem with the
Irreducibility Thesis is that it has to be reconciled with the Translatability Thesis,
according to which some imagistic thoughts can be translated into propositional
ones.
 2

What is thinking?

Although thinking is a familiar phenomenon, there is no full-fledged theory

of what thinking is. In philosophy and cognitive science, the issue of thinking
is rarely tackled directly. Even if some authors do try to tackle the problem of
thinking (e.g. Ryle, 1979), they usually deal with the problem indirectly, coming
out with telling descriptions of neighbouring matters, but rarely coming out
with a clear description of thinking alone.
It seems that any theory of thought should meet three requirements. First, it
should be able to deliver a theory of knowledge, for knowledge is a structure based
on thoughts. Second, a theory of thought should be able to deliver a theory of
content, for we want to know how to determine what we think of. Third, it should
be able to deliver a metaphysical theory of what the bearers of thoughts are.
These three requirements challenge any attempt to formulate a compelling
theory of imagistic thinking. The scepticism concerning the concept of imagistic
thinking is based on the belief that images do not have epistemological, semantic
and metaphysical features that we are willing to ascribe to thinking.
At the same time, a loose set of beliefs and intuitions about thinking builds
the so-called Received View of what thoughts and thinking are. The Received
View is not an actual theory. It is rather a research programme rooted in the
Kantian tradition. It consists of the belief that thinking is propositional – it is
based on logical structures made of true-evaluable components – and action-
based (Bayne, 2013).1 The strength of the Received View is that it seems to meet
the requirements for the theory of thinking. Moreover, it is in line with our
deeply rooted intuitions about the nature of thinking.
However, the Received View seems to contradict the imagistic theory
of thought. Thus, any attempt to create such a theory requires going beyond
the Received View, that is, broadening the notion of thinking so that it can
include imagistic thoughts, meeting the requirements of the theory of thoughts
simultaneously.
28 Thinking in Images

The Epistemological Challenge

Any theory of thinking should be able to deliver a theory of knowledge. For

although thinking and knowledge are separate phenomena, they are conceptually
bound. One of the main reasons we value thinking is that it is a knowledge-
building operation. It means that the main role of thinking is broadening our
knowledge. Similarly, knowledge is not a matter of coincidence; it is a product of
thoughtful actions – reasoning, deduction, hypotheses and so on. For all these
knowledge-building operations, thinking is indispensable.
One of the main reasons for the scepticism surrounding the concept of
imagistic thinking is the problem of its relation to knowledge. According to
the sceptical line of argument, an imagistic theory of thinking cannot deliver a
theory of knowledge, for images are not truth-bearers and do not have a logical
form.

The Epistemological Challenge: An imagistic theory of thinking cannot provide

a theory of knowledge.

The Epistemological Challenge does not hold that images cannot be a source
of knowledge. They obviously can. They can prompt some knowledge; they can
justify modal (e.g. Williamson, 2016) and factual knowledge (e.g. Dorsch, 2016;
Kind, 2018). Images can be direct objects of research. Images can be efficient
tools for thinking (e.g. Larkin and Simon, 1987; Sloman, 1978, 2002). They can
present procedural steps of reasoning in a more comprehensive manner and can
be part of argumentation processes. Thus, images can be a source of knowledge,
mainly because the human mind cannot grasp all information discursively.
However, knowledge does not consist of images. In the same fashion, perception
is a source of knowledge; and yet it is not an element building the structure
of knowledge. According to the Epistemological Challenge, images have an
instrumental nature – they help us acquire knowledge but do not constitute
knowledge.
It is held that knowledge consists in providing information that something is
the case. It is truth-evaluable and consists of elements that have truth-conditions.
To have truth-conditions means to be a truth-bearer. Propositions are believed
to be primary truth-bearers, which means that their content can be evaluated
according to whether they are true or false. Knowledge consists of truth-
evaluable and justifiable elements, such as propositions and logically structured
relations between these elements. Thus, the knowledge that something is the
case is traditionally dubbed propositional knowledge.
 What Is Thinking? 29

A propositional theory of thinking holds that thinking is a relation to

propositions building propositional attitudes. It gives us an answer to the
question of what the relation between thinking and knowledge is. If thinking
is a relation to a proposition, then it is self-explanatory that it can be part of
propositional knowledge. In contrast, the relationship between imagistic
thinking and propositional knowledge is debatable.
On the surface, images can be true too. Images can work as arguments in
scientific and legal practice. For instance, a photograph of a robber can serve as
evidence in court. There are also times when we want to say of an image ‘how
true!’ or assert that one image is closer to the facts than another (e.g. Eaton,
1980; Perini, 2005, 2012; Solt, 1989; Zeimbekis, 2015).
These intuitions, however, are clearly inconclusive. For to function in an
argumentation process, an object does not have to be veridical. For instance, a
knife can serve as evidence in court, but that does not mean that the knife is true.
The knife makes the argument true, but it is not true in itself. In other words,
images can function as truth-makers, but they do not have to be truth-bearers
(e.g. Goodwin, 2009; Heck, 2007).
Moreover, one can posit that images are not truth-evaluable and that they
can still provide true descriptions. Based on John’s portrait representing
him as bald, we can describe him as bald. It does not mean, however, that
images are true or false. Similarly, a knife at the crime scene can provide a true
description of what has happened. Yet it does not mean that the knife is true.
We often say that photographs do not lie, yet this concept of truth therein is
not the same as in the case of propositions. Images can be used to form true
descriptions, but that is not to say that images are true or false in the same
sense as descriptions are.
Still, we need an argument for why images are not truth-evaluable and cannot
deliver a theory of knowledge. We also need to know what kind of argument we
do not want. For one thing, one can argue that iconic content can be accurate or
inaccurate (Crane, 2009; Greenberg, 2018, 2021). Accuracy comes in degrees;
truth does not. A picture of John can be more or less accurate but not true or
false. A proposition cannot be more or less true. Truth and falsehood are all or
nothing. This argument, however, is unsound, for we can easily apply the idea of
coming in degrees to a set of complex propositions.
For another, one can argue that iconic content is informatively rich (Dretske,
1981; Kitcher and Varzi, 2000) and finely grained (Tye, 2005). It means, first,
that iconic content conveys too much information to express it with a finite set
of propositions and, second, that iconic content is detailed and determined. In
30 Thinking in Images

contrast, propositional content is general and abstract. This argument, however,

is unsound. There are propositions that are rich in content, such as ‘π equals
3,14 . . .’, and images that are informatively primitive, such as an image of a dot.
Further, propositional content can be more fine-grained than iconic content.
Pictures of aqua and cyan objects are often not detailed enough to see the
difference between them – the propositions ‘x is aqua’ and ‘y is cyan’ are.
In the next three sections, I present two positive arguments for a non-
propositional nature of images: Wittgenstein’s argument from content
indeterminacy and Frege-Davidson’s argument from lack of logical form. Both
arguments will be the basis of semantical and metaphysical objections as well.
Therefore, they will be presented here in a detailed manner.

Wittgenstein’s argument from content indeterminacy

Wittgenstein’s scepticism regarding the truth-evaluability of images concerns

mainly the indeterminacy of iconic content. The argument is illustrated by a well-
known remark from Philosophical Investigations (1953, §139): ‘I see a picture; it
represents an old man walking up a steep path leaning on a stick.—How? Might
it not have looked just the same if he had been sliding downhill in that position?’
Wittgenstein holds that the same image can have various interpretations,
depending on how it is taken. The walking-man image certainly represents
some colour and spatial relations. But does it represent the man walking up or
sliding down? In the case of language, you can determine the cognitive meaning
of a sentence based on the meaning of its constituents and syntactic rules. The
content of images is indeterminable.
The indeterminacy argument can be easily extended. Images cannot
distinguish intensional, self-referential or modal contexts. A walking-man
image can mean either that I believe that the man is walking, that I doubt that
he is walking or that I hope that he is walking. The same image can express the
thought that a man is walking and that a bank-owner is walking. It expresses
the thought the man is walking, as well as the self-referential thought I think
of the thought that he is walking. It may mean that it is possible that
he is walking or that it is necessary. Images cannot distinguish these contexts;
language can.
Let me clarify the matter. Wittgenstein's argument is not that images
are ambiguous. Language is ambiguous, too. We can never know what a
 What Is Thinking? 31

sentence means in a communicative context. However, we can assign different

interpretations of the sentence to different propositions, which are not
ambiguous. Wittgenstein’s argument is metaphysical, not epistemological.2 He
argues that the meaning of images is indeterminable because there is always a
method of projection that results in a certain image (Wittgenstein, 1958). To
determine which method of projection is relevant to interpreting the content of
an image, one has to know what the image represents. However, that is what was
supposed to be determined by the interpretation. Thus, there can be no correct
interpretation of the content of images, for the concept of correctness does not
apply to it.
In consequence, images are insufficient to determine their truth-conditions
and therefore are not truth-bearers; for to hold that representation is true,
we need a systematic way to isolate the representational content and state the
fact the representation refers to. Tarski (1956) accomplished that for formal
representational systems by distinguishing the sequence of atomic symbols
of the system and providing the means to translate them into metalanguage.
Granted that Wittgenstein’s argument holds, the same cannot be done for
imagistic representational systems. If there is no systematic representation–
reference relationship, then the concept of truth is not applicable.
Moreover, as Wittgenstein underlines, any way to counter this argument
referring to some internal relation fixing images’ content, such as inner knowledge
of the meaning, makes this content inevitably private and vulnerable to private-
language argument. If I knew what an image means, thanks to some internal
feeling of knowing the meaning, then such knowledge would be unverifiable.
Kripke (1982, 51) rightly calls this strategy ‘mysterious and desperate’.
How should one evaluate the indeterminacy argument? Fodor calls it ‘entirely
convincing’ (1975, 180) and uses it to argue that images cannot constitute
thoughts. According to Fodor, to fix the content of an image, one has to name or
describe it. For instance, titling a picture ‘the walking-up-man’ determines that
the picture represents a man as walking up.
Yet the scope of the indeterminacy argument is limited (e.g. Goldberg,
2017). Wittgenstein’s point is not that images are not bearers of thoughts and
propositions are. For one thing, he puts too much effort into elaborating a
pictorial way of describing ideas – in his works, he uses approximately 1,300
different images to present philosophical problems (e.g. Nyíri, 2013; Richtmeyer,
2019). For another, if images were deprived of meaning, then it would be
senseless to speak of different uses of the same picture of an old man. In other
words, one has to distinguish between the picture and its uses, in the same way
32 Thinking in Images

as we distinguish between a rule and an application of the rule (Egan, 2011;

Roser, 1996).
Thus, the goal of the argument is not to argue that we cannot think with
images but that the role we attribute to images is misconceived. In other words,
it is not the case that images cannot be bearers of thoughts but that no mental
structure can play the role we attribute to images.
Wittgenstein’s argument is directed against Russell’s theory of image-
propositions (List, 1981; Russell, 1919, 1921), according to which propositions
can have either word form or imagistic form. The meanings of images are
primitive – we know them directly and without reflection – and determined
by the resemblance to the represented object. Accordingly, images are true if
and only if there is a function mapping the image onto reality. For instance, a
memory image representing John as bald is true if and only if John is bald.
According to Wittgenstein, if images were to serve a propositional-like
role and their meaning were known directly based on perceiving similarities,
then their contents would be indeterminable and could not be veridical. In
particular, the resemblance relation cannot be epistemically primitive, for
in order to know what aspect of resemblance is relevant for representation,
one must know what is represented (Goodman, 1972, 1976; Suárez, 2003;
Wollheim, 1980, 1987, 1998).
Yet, Wittgenstein’s argument shows that images cannot play the same role as
propositions, for we cannot determine the meaning of images based solely on
resemblance relations. Moreover, the relation between images and thoughts is
not analogous to the relation between language and thoughts. Images cannot
play the same role as sentences. However, it does not support the claim that
images play no role in thinking.
On the surface, two argumentative strategies can help us avoid Wittgenstein’s
conclusion. Both of them, however, fail. First, it is tempting to address
Wittgenstein’s argument by holding that we never approach images with no
fixed preferences. After all, images are part of the communication process (e.g.
Abell, 2009; Bantinaki, 2007; Blumson, 2009, 2014; Frixione and Lombardii,
2015). Therefore, they are never fully indeterminate. Some interpretations
are more likely than others based on communicative context or knowing the
author’s intentions. For instance, Abell (2009), drawing on Gricean theory of
meaning, holds that relevant aspects of resemblance and the content of a picture
can be determined by knowing the author’s communicative intentions. In short,
a picture resembles an object O in the relevant respect if an author intends
the picture to resemble O in this respect to bring O to the viewer’s mind and
 What Is Thinking? 33

intends that this resemblance have this effect because the viewer recognizes this
intention.
However, this strategy is a dead end. We can appeal to the communicative
context in the case of pictures. This strategy fails in the case of mental images.
Internal representations are not part of any communicative practice.
We cannot determine the content of images based on the author’s intentions,
either. It would be circular to appeal to the author’s intentions to determine the
content of mental image. To know our own intentions, we need to know the
content of mental images, but to know the content of mental images, we would
need to know the content of our intentions. It would not be a problem if we
could easily distinguish between the content of mental images and pictures. Yet,
that would not provide a unified account of imagistic content.
Moreover, the intention-based accounts of depiction cannot explain how
pictures represent. X-ray pictures are correct or incorrect regardless of the
intentions of the author. Automatically taken pictures have content but not
authors. There are unfulfilled intentions too. If I intend to draw a horse, but
due to a lack of skill, the drawing is more like a cow, we would not say that the
drawing represents a horse only because I intended to depict one. Therefore,
the communicative context and author’s intentions cannot solve the problem of
content indeterminacy.
Second, we can claim that although pictorial content is indeterminate, it is not
indeterminate all the way down. At least part of the content is determined. We
can claim that there is some primitive content that is fixed and constrains possible
interpretations of pictures. Kulvicki (2006a, 2014, 2020), following Haugeland
(1998), calls it bare bones content. Bare bones content is defined in terms of
projective invariants, that is, these representational features that are common for
different interpretations of a picture, such as certain patterns of colours or spatial
relations.3 Bare bones content goes beyond the context of interpretation and
captures basic semantic information displayed in pictures. It can be compared
to the concept of character in the philosophy of language (Kaplan, 1989). In
contrast, fleshed-out content is content that we can describe as a picture of John
or a map of London and depends on our ability to recognize kinds of objects by
the features of bare bones content. It can be compared to Kaplan’s concept of
content. According to Kulvicki (2006a), although pictures admit of alternative
interpretations of the fleshed-out content, all these interpretations have a
common bare bones content that constrains possible interpretations of the
picture. For example, a square-like picture can represent a square or a face of the
cube seen from a certain angle. These different interpretations mark the fleshed-
34 Thinking in Images

out content. Yet the square-like shape is invariant in different interpretational

contexts. It forms the bare bones content and constraints possible interpretations,
for example a square-like shape cannot represent a circle.
This strategy, however, won’t do. Primarily, we must determine which features
are relevant as bare bones content. Some features, such as being painted on a
canvas, are projective invariants but do not constitute bare bones content. Thus,
to know which features are relevant, we must determine the class of abstraction.
However, in line with Wittgenstein’s argument, we need to know what the
picture’s fleshed-out content is to determine the class of abstraction. Consider
two examples. When interpreting a topological map of the underground, we
must abstract away from spatial distances. To do that, however, we need to know
that it is a topological and not a topographic map, which means that we have to
know what the map represents – either topological or topographical relations.
Next, in the case of a black-and-white picture, we abstract away from the
patterns of colorus. We are justified in doing so only if we know what the picture
represents – is it a black-and-white picture of a coloured object or a coloured
picture of a black-and-white object, such as a penguin? Thus, bare bones content
cannot constrain possible interpretations of fleshed-out content, for to know
what the bare bones content is, we need to determine the fleshed-out content.
Second of all, even if we acknowledge that bare bones content can be
independently determined, it does not limit possible interpretations. We need
to know more than just how to distinguish between accurate but different
interpretations of given content. We need a differentiating criterion between
accurate and inaccurate representations too. Obviously, there are inaccurate
pictures. The question is how to distinguish them from accurate ones. For
example, the same map can accurately represent London but also inaccurately
represent Berlin. After all, having a map does not mean that we have a good map.
However, the bare bones content of an accurate map of London and an inaccurate
map of Berlin is the same. Thus, the indeterminacy of content is unavoidable.

Frege-Davidson’s argument from lack of logical form

Frege’s argument against imagistic thinking is based on images’ inability to

form logical relations. Logical relations are applicable only to truth-bearers.
Propositions can be true or false, and the truth is the central semantic concept
of propositional logic. As Frege (1984, 351) puts it: ‘Just as “beautiful” points the
way for aesthetics and “good” for ethics, so do words like “true” for logic.’
 What Is Thinking? 35

For logical relations to hold, the elements of the relation have to possess
a logical form, that is, a syntactically fixed structure, such as a set of logical
constants and variables, with determined transformational rules that preserve the
logical values of its components. If A implies B, then based on transformational
rules, it is possible to transform the truth of the first into the truth of the second.
Propositional logic shows how the truth and falsehood of complex propositions
depend on the truth and falsehood of simple ones. Truth-functions operate on
propositions that can be negated, disjointed or conjoined; they can imply one
another or be equivalent. One of the main reasons for talking about propositions
at all is that they explain how things can stand in these logical relations.
Images lack logical form. They can illustrate logical relations, just like
Venn diagrams can illustrate relations between sets, and apples can illustrate
calculations. Images can help us to grasp logical transitions, just like apples can
help us to grasp calculations. Yet they do not have a logical structure, just like
apples are not logical entities. There are no truth-preserving transformation
rules for imagistic representation. There is no pictorial negation (Crane, 2009;
Sainsbury, 2005), conjunction or disjunction (Heck, 2007); images cannot
express implications, logical modalities, quantifications (Frege, 1984) and so on.
For instance, there is no image of all people being bald or such that it is necessary
that they are bald.
Language is essential here. Grammatical constructions indicate relations
between concepts and point to contradictions, consequences and correspondence
between pieces of information (Loewenstein and Gentner, 2005; Lupyan et al.,
2007). Images do not provide such opportunities.
Thus, images do not do any logical work. The best explanation of these facts
is that images are not truth-bearers the transformational rules can be applied to.
According to Frege, the logical structure of thoughts mirrors the logical
structure of language. Thus, the only route to an analysis of the logical structure
of thought goes through an analysis of language. Dummett (1993, 128) calls it
‘the fundamental axiom of analytical philosophy’.
The implication of Frege’s argument is that it draws a clear line between images
and beliefs. Beliefs are interpretable in logical terms; images are not. Beliefs are
‘inferentially promiscuous’ (Stich, 1978), which means that they can figure as
premises in inferential transitions. It helps us to explain why images behave
differently than beliefs. Images are not committed to truth; beliefs are. I cannot
believe that 2 > 3 if I know that it is the case that 3 > 2. In contrast, I know that it
is the case that there can be no flying horses, and yet that is not a reason to refrain
from depicting one. Deprived of logical form, images are independent of beliefs.
36 Thinking in Images

We find a similar argument in Davidson (1997, 2001). According to him,

thinking differs from other mental states in that it can be rational. A general
theory of thinking should explain how it is possible to think rationally.
Rationality is a matter of standing in logical relations. Beliefs are logically linked
to each other in the form of a ‘web of belief ’. They justify each other and can
rationalize one’s actions. The only way to express these rational relations is
through language. That is the main reason why Davidson holds that language is
necessary for thinking (Davidson, 1982).
In contrast, lack of logical form renders images a-rational; they are neither
rational nor irrational – the concept of rationality simply does not apply to them.
If we think of images in terms of resemblance, then there is no room for asking
about reasons. There can be questions about reasons when stating that p is q, but
there cannot be a question of reasons when a picture depicts p as q. You can only
show that, not argue it (Mersch, 2011).
Relations between images are not logical; these relations are usually
understood as a causal chain of associations. The image of a mother can evoke
a memory image of a family home, but the link between these two images is not
a matter of logical consequence. We can speak of images’ temporal or causal
sequence, but rationality is not based on temporal or causal links. It is a matter
of following logical rules.
Therefore, images cannot be rational. The difficulty is not that all thoughts
are rational. They are certainly not. Some thoughts are irrational. However, if
the concept of rationality does not apply to images, imagistic thinking is neither
rational nor irrational. Thus, images cannot be constituents of thoughts.

The pictorial fallacy

At first glance, it may seem that the Frege-Davidson argument can be refuted by
pointing out straightforward counterexamples. For instance, if I want to negate
that John has red hair, I can depict him with blond hair. If I depict a green and
a red apple, I express an alternative of a green and red apple. If one places two
pictures next to each other, much like in a comic book, then one can say that
their content is conjoined or that one implies the other (e.g. Malinas, 1991;
Westerhoff, 2005).
These examples are misleading. The role of logical form is to determine the
truth-conditions of its elements. In the case of images, truth-conditions cannot
be determined. Having a picture of John with blond hair may be a negation of
 What Is Thinking? 37

him having red hair, or black hair and so on; for the content of not having red hair
is not simply being blond but an infinite alternative of the form ‘having blond
hair, or having black hair, or having green hair, etc.’. No image can represent
infinitely many properties.
By the same token, conjunction, disjunction and implication do not simply
represent a sequence of its elements; they set up a logical link between their
components. In the case of two pictures, there is no way to determine the nature
of this link – whether it is a temporal sequence, causal link, spatial transformation
or simply a set of two unrelated pictures. In all these cases, the pictorial form is
the same.
Moreover, images lack the generality and precision required by logical
operations. For instance, it is impossible to depict the difference between the
claim that ∃xP(x) and ∀xP(x) (Hintikka, 1987).
However, there is a more sophisticated line of argument available. It is based
on two points. First, opponents of the propositional view of iconic content, such
as Crane, usually agree that we can negate pictorial content. For instance, we
can point at a picture representing John as bald and claim ‘it is not the case that
John is bald’. According to Grzankowski (2015, 2018), negation is a good way of
testing for propositional content. If something can be negated, then it is truth-
evaluable. Second, when seeing a picture, we can describe the picture’s content in
terms of modal propositions. For instance, when pointing at a picture depicting
John being bald, we can assert that it indicates that in some possible world, it
is true that John is bald (e.g. Kulvicki, 2020; Malinas, 1991; Matthen, 2014).
Consequently, if the content can be negated and if we can say of the picture’s
content that it represents logical modalities, then it has a logical form.
This argument is invalid. Yet, it is invalid in an instructive way. Logical
operators apply to propositions. They do not apply to the way pictures look.
Thus, the picture’s content must be put into a propositional form to which we
attribute a logical operator. This is clearly possible. As Kaplan (1968) observed,
many of our beliefs are of a form such as ‘the colour of her eyes is [_]’, where the
blank ‘[_]’ is filled with a pictorial representation. Hence, images can figure in
what we believe and can play a role in forming propositional contexts.
Yet, the fact that something can be put into a propositional context does not
imply that it has propositional content. Pointing at a knife can be used to express
the proposition that ‘it is (not) possible that it is the weapon Hamlet killed
Polonius with’, where ‘it’ is a deictic word, the referent of which is determined by
indication. It does not mean that the knife is propositional. The proposition has
been expressed deictically by indicating the knife. Still, the propositional content
38 Thinking in Images

characterizes the deictic use of the word ‘it’ and an ostensive act of indicating the
knife – not the knife as such.
This context can be multiplied, as in the case of deferred reference (Nunberg,
1993; Quine, 1968). Deferred reference is the use of an expression to refer to an
entity that is not denoted directly by this expression. For instance, I can point at
Quine’s photograph and say that ‘that is most probably the greatest philosopher
of the twentieth century’. The word ‘that’ does not refer to the photograph but
to Quine. The visual properties of the photograph are used to recognize Quine
or, putting it differently, to transfer the reference from the photograph to Quine.
Similarly, I can point at John’s portrait and express the proposition that it is
possible that he is bald. Yet, the propositional content characterizes the deictic
use of the picture, not the picture itself. The picture is used here in a deferred
context. We represent John as bald to transfer the property of baldness to John.
Properties of representation are used to transfer reference to the represented
object. In our example, the first attribution – representing John as bald – concerns
the properties of representation and the second – attributing baldness to John
– the properties of the represented object. Properties of representation describe
the way the object is represented. Properties of represented objects describe the
object itself. Only when we attribute properties to the represented objects can we
speak of truth-conditions. That is why I can depict John as bald without being
committed to holding the belief that John is bald.
Thus, we have to distinguish between the properties of representation and
the properties of the represented objects. Properties of representation single out
the way something can be represented. Properties of represented objects single
out the way an object is or can be. The inability to distinguish these two kinds of
properties is only another instantiation of the phenomenological fallacy. Let us
dub it the ‘pictorial fallacy’.
Let me illustrate the pictorial fallacy with the help of a drawing of a flying
horse. If I draw a flying horse, I can attribute the ability of flight to the depiction
of the horse. Therefore, I can say that it is possible to represent horses as flying,
which is true. Yet it is something different than saying that it is possible that
horses can fly, which is necessarily false in the same way as saying that it is
possible that water is not H2O.4
The argument that images can express modal and negative propositions is an
instantiation of the pictorial fallacy. Note that the force of this argument depends
on how we understand the concept of ‘expression’. On the one hand, we can say
that numerals express numbers, which means that numerals are necessary means
to represent numbers. On the other, we can say that a book cover expresses the
 What Is Thinking? 39

content of a book, which means that the book cover is an auxiliary means to
understand the content of the book. The properties of the book cover represent
the book’s content and are used to recognize the properties of the book’s content.
In other words, the book cover is an illustration of the book’s content. However,
the book does not need illustrations to have content.
Pictures can illustrate modal and negative propositions. They can represent
something as possible and negated, but they do not represent that something
can be or is not the case; propositions do. Thus, indicating the fact that pictures
can express modal and negative propositions does not imply that pictures have
propositional content and can stand in logical relations.

The Semantical Challenge

According to well-established philosophical tradition, thoughts are intentional;

they can stand for things, properties and states of affairs. In other words, they
have content. Any theory of thinking should be able to deliver a satisfactory
theory of content.
According to the same tradition, images are intentional, for they can
represent the world by resembling it. Let us call it the Traditional View on
the nature of depiction. According to the Traditional View, a picture of a red
square stands for a red square for it resembles a red square. The problem with
such a formulation of the intentionality of pictures is that if it were true, then
an imagistic theory of thinking could not provide a theory of the content of
thoughts.

The Semantical Challenge: An imagistic theory of thinking cannot provide a

theory of the content of thoughts.

Frege-Davidson’s argument from lack of logical form and Wittgenstein’s

argument from content indeterminacy form the backbone of the Semantical
Challenge. Both arguments imply that images cannot determine the content of
thoughts and express predicative functions.5 It shows that if the standard views
of intentionality and pictorial semantics are true, then the content of imagistic
thoughts is parasitic upon the existence of some more fundamental systems of
representation.
One of the distinguishing features of thoughts is being intentional in de re
sense. De re intentionality is most often compared to de dicto intentionality.
Formally speaking, this distinction concerns ranges of modal operators; either
40 Thinking in Images

a is such that it is possible that it is P or that it is possible that a is P. Different

interpretations of the following expression can illustrate it:

(1) John believes that someone is following him.

According to the de dicto interpretation, John believes someone is following him.

According to the de re interpretation, there is someone that John believes is following
him (Quine, 1956). The truth of the de dicto interpretation of (1) would give a
psychiatrist reason to be interested in John. The truth of the de re interpretation
would give the police reason to investigate the case, for, unlike de dicto attitudes, de
re attitudes attribute properties directly to the object of the attitude.
The different interpretations of (1) illustrate that having a de dicto attitude
is insufficient for having a de re attitude. In other words, the belief that a
proposition containing a term referring to something is true is not sufficient for
having a belief about the object to which the term refers. The transition from de
dicto to de re intentionality requires a special cognitive relation to the object (see
Kaplan, 1968).
How should one think about this cognitive relation? Russell’s idea (1997)
was that the intentionality of thoughts (in the de re interpretation) requires that
one has to know what one is thinking of. Evans called it Russell’s Principle and
expressed it as follows: ‘a subject cannot make a judgement about something
unless he knows which object his judgement is about’ (Evans, 1982, 89).
According to Russell’s Principle, intentionality de re is connected to
knowledge of the object one attributes properties to. In other words, one cannot
have thoughts of an object unless one possesses knowledge about the object
one thinks about. Evans describes this knowledge in terms of discriminating
capacities and dubs it ‘discriminating knowledge’. The idea is that if one ascribes
the belief Fido is a dog to a cognitive system S, then it should be possible to
ascribe to S a capacity to recognize the object the property being a dog can be
attributed to. Only if we can discriminate the object of our belief, namely Fido,
can we predicate of the object Fido that it is a dog, which can be expressed with
the sentence ‘Fido is a dog’.
It does not imply that thinking that Fido is a dog is equivalent to having
knowledge that Fido is a dog or that the act of thinking is equivalent to the acts
of knowledge and cognition – one can think whatever one wants. It does not
imply that the thought cannot be wrong – Fido could be a cat. It does not mean
that we cannot fail in the ascription of thoughts either. As Freud might say, it
may seem to us that we are thinking of a dog while we are actually thinking of
our mother.
 What Is Thinking? 41

However, possessing discriminating knowledge means that having the

thought Fido is a dog entails a capacity to recognize the object one is thinking
about – something which is necessary to report that one is thinking of Fido and
not about anything else. Without such knowledge, there would be no object to
predicate of and therefore there would be no thought at all.
Evans does not provide an explication of the principle. Rather, he leaves us
with an example of a case where this principle is not satisfied. Suppose that a
subject sees two steel balls continuously rotating about some point. At a later
time, he reminisces that shiny ball is pretty, but he does not know that in the
meantime, he experienced amnesia that caused him to forget about one of the
balls. In such a case, Evans holds that the subject does not have discriminating
knowledge of the object he is thinking of; therefore, he has no beliefs about it, for
he does not know which object can be described as pretty. As Evans states (1982,
115), ‘there is nothing else he can do which will show that his thought is really
about one of the two balls (that ball) rather than the other.’
Evans’s position is definitely very strong and vulnerable to counterexamples
(e.g. Hawthorne and Manley, 2012; Rozemond, 1993). For instance, I can think
of the girl I saw yesterday in the park, even though I cannot distinguish the girl
from her twin sister I had seen two days ago, not knowing which of the twin
sisters I am thinking of. Even if I forget that I saw one of them two days ago, the
name ‘the girl I saw yesterday’ fixes the reference of my thought. The idea is that
we can pick out the referent and determine the content of thought without being
able to discern the sisters; whichever sister I am thinking of, I am only thinking
of the one I saw yesterday and not the one I had seen two days ago. I can be
wrong in connecting this thought to one of the twin sisters in particular, but that
does not mean I have no thought.
What does this counterexample show? It shows either that Russell’s Principle
is invalid or that Evans’s argument in favour of the principle is invalid. However,
we want to keep Russell’s Principle for many independent reasons. For one
thing, to veridically apply a predicate to an object, we must be able to identify
the object (e.g. Burge, 2010). For another, to infer the properties of an object,
we must be able to establish that the inference relates to the same object (e.g.
Gersel, 2017), in which case, something must have gone wrong with the steel
ball example.
Note that the main idea of Russell’s Principle is that the object we predicate
of can be individuated. The steel ball image, says Evans, does not allow us to
individuate the referent. That, however, says something important about the
nature of an image and its relation to thought and not about Russell’s Principle
42 Thinking in Images

itself. Although we have the conceptual capacities to individuate the steel balls –
for instance, with the concepts this ball and not-this ball – we may lack the
skill to apply these concepts successfully. We may fail in concept-application, but
that is not particularly unexpected. I may think that the girl I saw yesterday was
Jane, but it was Mary – her twin sister. It may happen.
In contrast, images do not provide us with tools to individuate objects we
predicate of and it is debatable whether they can express predication. First, as
Wittgenstein’s argument shows, the content of images cannot be independently
determined. The problem is not that we cannot be wrong about the interpretation
of the content of an imagistic thought. The difficulty is that if the content of
the images is indeterminable, then there can be no correct interpretation of the
content. Thus, images cannot satisfy Russell’s Principle. Propositions can.
Second, as Frege-Davidson’s argument indicates, images lack logical form. As
a consequence, we face the so-called binding problem of iconic representations.
It is analogous to the unity of propositions problem in the philosophy of
language (e.g. Gaskin, 2008). In the philosophy of language, the question is how
to distinguish a proposition, such as this square is red, from a list of names
‘this square’, ‘is’ and ‘red’. The answer comes easier if we assume that propositions
have a logical structure that puts the names together. In the case of an image of a
red square, it is either an image of a square, where the colour is irrelevant, such
as in the case of black-and-white photographs, or an image of redness, where
the shape is irrelevant, such as in the case of colour samples or, lastly, an image
of a red square. The problem is that without a logical structure, it cannot be
determined whether the content is combined or not.
Consequently, images are not capable of expressing predication, for
predication requires a logical apparatus they lack. In particular, it requires a
separation between arguments and predications. Propositional representations
can do that. In the sentence ‘it is a red square’, we can point out the part that
represents the argument and the part that corresponds to the predicate. In an
image of a red square, the same part of the picture corresponds to the argument
and the predicate. Consequently, a proposition that a square is red can predicate
the redness of the square, for the square takes the place of a named argument
we predicate redness of. In contrast, an image of a red square is in line with the
interpretation that the square is red and that the colour red is square-like.
Let me clarify this point. Images surely bind contents together. An image of a
red square combines red and square and differs semantically from an image of
a red triangle or a green square. However, the Semantical Challenge shows that
we cannot determine the content of imagistic thoughts.
 What Is Thinking? 43

What I am suggesting here is that the Traditional View makes them vulnerable
to the Semantical Challenge. That may be, however, an argument for changing
our traditional way of thinking about images and not against an imagistic theory
of thought.

The Metaphysical Challenge

It is believed (e.g., Davidson, 1997; Laurence and Margolis, 2012; Solomon et al.,
1999) that concepts are the building blocks of thoughts. Why is that so?
As Frege-Davidson’s argument shows, one of the distinctive features of
thoughts is that they are systematically structured. Entertaining a thought
of one kind entails a capacity to entertain a thought of another kind. For
instance, entertaining the thought that John is happy and that Mary is sad is
systematically connected with the cognitive ability to entertain the thought that
Mary is happy and that John is sad. Having the thought that John is happy
entails a capacity to think that someone is happy. In Evans’s words (1982, 104),
‘if a subject can be credited with the thought that a is F, then he must have the
conceptual resources for entertaining the thought that a is G, for every property
of being G of which he has a conception.’
The same rule applies to inferences (e.g. Fodor and Pylyshyn, 1988). If I think
it is dark and cold and raining, I can infer that it is cold and raining; for
from P & Q I can infer that P (or Q). By the same token, I have to be capable of
inferring from it is cold and raining that it is raining. If I cannot do so, I
do not know what inference is.
Thus, thoughts must be systematically co-related (e.g. Heck, 2000; Peacocke,
1992), which means that they are systematic in nature. Evans (1982) calls this
requirement the Generality Constraint.
To meet this requirement, thoughts have to consist of recombinable
constituents that can build more complex structures. It means that thoughts
are compositional in nature. The compositionality of thoughts means, first,
that the meaning of a complex thought is determined by the meaning of
its constituents. The constituents of thoughts are parts of thoughts that are
canonically distinguishable, for not every partition of thought makes sense. The
idea is that canonically decomposed parts are syntactically and semantically
meaningful units. For instance, the thought that John loves Mary can be
decomposed into John loves and Mary, but not into John . . . Mary (e.g.
Fodor, 2008).
44 Thinking in Images

Second, the meaning of complex thoughts must come from the meaning
of its canonically distinguishable parts together with the rules of composition,
for not all combinations are allowed. The recombination of the parts must be
meaningful. John loves Mary can be recombined into Mary loves John
but not into John Mary loves.6 These rules are recursive. If I have a thought
that John loves his mother, I have to be capable of having a thought that
John loves his mother’s mother and so on. Putting it together, it means
that thoughts have a recursive syntax that combines canonically distinguishable
parts according to the combinatorial rules (e.g. Pagin and Westerståhl, 2010).
Systematicity and compositionality allow us to explain how thoughts can
be productive. The productivity of thoughts means that we can entertain a
potentially infinite number of different thoughts. Moreover, we can understand
an indeterminate number of thoughts we have never entertained before. It all
makes sense if we assume that thoughts are made of systematically structured
and combinable elements. These elements are concepts, and the complex
conceptual structures are propositions (Peacocke, 1992).
Language seems to be systematic and compositional. It has syntax and
distinguishable syntactic and semantic parts. Thus, one can hold that thoughts
consist of language-like representations (e.g. Devitt, 2006). In contrast, images
lack systematicity and compositionality; therefore, they do not have metaphysical
properties we are willing to ascribe to thoughts.

The Metaphysical Challenge: an imagistic theory of thinking cannot provide a

theory accounting for systematicity and compositionality of thoughts.

Images are neither systematic nor compositional. As Fodor shows, they lack
syntactic structure and canonical decomposition.7 Images lack grammar capable
of generating infinite sequences of sentences (Eco, 1995; Wollheim, 1993).
Therefore, iconic representations do not meet the Generality Constraint.
Fodor’s argument (2007, 2008) takes the form of the so-called Picture
Principle. According to the Picture Principle, iconic representations can be
distinguished topologically: although images have interpretable parts, they lack
canonical decomposition. It means, loosely, that we can cut up an image however
we like, and each image-part will represent a relevant part of the represented
object. Thus, every part of the representation represents some part of the scene
represented by the whole representation (e.g. Green and Quilty-Dunn, 2021;
Quilty-Dunn, 2016, 2020; Sober, 1976). In contrast, discursive representations
have canonical decompositions, which means that they cannot be cut into
pieces however we like. Discursive representations have constituent parts. For
 What Is Thinking? 45

instance, the content of the proposition snow is white can be decomposed

into the parts snow and is white but not into snow . . . white, which means
that the expression ‘snow . . . white’ does not possess independent semantical
value. Thus, although images can be decomposed, they cannot be canonically
decomposed. However, if they can be composed and decomposed however one
wants, then they lack syntactical structure.
Two clarifications are needed. First, according to the Metaphysical Challenge,
images can still be useful in thinking. They can prompt some thoughts or
facilitate information processing. However, they cannot be bearers of thoughts,
for bearers of thoughts must meet the requirements of systematicity and
compositionality (Pylyshyn, 2003a). Propositions can meet these requirements;
therefore, thinking has a propositional structure. Thinking with images can at
most be built upon operations carried out with propositions; therefore, images
are epiphenomena of propositional representations.
Second, one might object that images can exhibit systematicity. For one
thing, map-like representations seem to be systematic (e.g. Braddon-Mitchell
and Jackson, 1996; Camp, 2007). For example, a part of a map representing
that London is west of Berlin also represents that Berlin is east of London. For
another, as Matthen (2005) notes in criticizing Evans’s Generality Constraint, if
one is able to imagine a blue circle and a red square, then one is able to imagine
a red circle and a blue square. In other words, if a representational system can
represent multiple features together, it can represent different configurations of
these features.
These objections miss the mark. The systematicity of thoughts comes paired
with compositionality; for thoughts to be systematic, we have to be able to
distinguish between the meanings of the constituents and the meanings of the
complex structures. The thought John loves Mary is built out of the concepts
John, Mary and love, which can be distinguished as separate semantical
units. In the case of images, such a separation cannot be produced, for they lack
canonical decomposition.
Let us recall Wittgenstein’s indeterminacy argument. According to
Wittgenstein, the content of images cannot be independently determined.
The metaphysical underpinning of Wittgenstein’s argument is the observation
that imagistic information lacks the syntactic structure that binds together the
represented feature and the vehicle of representation.
Let me illustrate this claim. An image of a red square can represent the concept
of redness or squareness, as well as the proposition some squares are red. In
other words, if one has a mental image of a red square, does it express the singular
46 Thinking in Images

concept red square or the proposition some squares are red? Language-like
representations have syntactical structure and can distinguish between concepts
and propositional structures made of concepts. Images do not.
To sum up, images lack syntactic structure, and, therefore, they do not meet
the requirements of the Generality Constraint. They are neither systematic nor
compositional. Any theory of imagistic thinking has to address the problem of
systematicity and compositionality of thoughts.

The Received View

The epistemological, semantical and metaphysical challenges are met by the

so-called Received View, according to which thinking is propositional and
action-based. Moreover, the Received View seems to have more explanatory
power than competing theories.
The propositional theory of thought stems from the rationalist tradition
in philosophy. It is a set of positions that have in common the belief that the
feature defining human thinking is rationality and that thinking itself can be
modelled according to truth-preserving rules of logic (Fodor, 1991). Thinking is
a set of operations on propositions and a matter of holding certain relations to
propositions expressed in propositional attitude reports. Accordingly, thoughts
are manifestations of propositional attitudes. Propositionalism in twentieth-
century philosophy and cognitive science is represented by, among others,
Frege’s theory of thought, Fodor’s LOT, Newell and Simon’s GPS and mental
logic theories.
Frege’s theory of thought best expresses propositional theory. According to
Frege (1984), thinking is defined as a subject’s relation to thoughts. Thoughts are
most often interpreted as abstract objects we refer to by means of representations
of the type ‘that-p’ expressed in propositional attitude reports such as ‘I believe
that p’ or ‘I hope that p’. Thoughts are also distinct from the sentences that
express them. Different sentences can express the same thought, such as in the
expressions ‘snow is white’ and ‘Schnee ist weiß’. The constituents of thoughts
are concepts; in the same way, words are constituents of sentences expressing
thoughts. Thus, someone who thinks that Venus is the second planet from the
Sun must possess the concepts of Venus, planet, sun and 2.
Propositional theories are most often contrasted with associationist theories
of thought.8 The common root of associationism is the belief that rationality is
not the best way to describe relations between thoughts. Thoughts are related to
 What Is Thinking? 47

each other because, in their causal history, certain facts have causally linked the
mental states into pairs, ensuring that if one pair member is activated, the next
one is also activated. For instance, the frequency with which a certain organism
came into contact with events X and Y determines the frequency with which that
organism will have related thoughts about X and Y.
Hume’s theory of ideas illustrates this well. It is primarily a theory of how
perception (impressions) determines strings of thoughts (ideas). According to
Hume (1975), ideas are copies of impressions in the following fashion: if the
impressions IM1 and IM2 are related in perception, then the corresponding ideas
ID1 and ID2 are also related. We do not need to refer to any intermediary entities
like implicit rules. The causal order of perception thus determines the order of
thought.
Associationism has many advantages. It explains the psychological
mechanism of thought acquisition well. It is metaphysically unencumbered and
does not postulate any hidden logical mechanisms. It explains many mental
phenomena too. It is used in learning theory (e.g. behaviourism), theories of
reasoning (e.g. dual-process theories) and theories of thought implementation
(e.g. connectionism). It is in line with nowadays theories of mind, such as
Enactivism and Dynamic Systems theories. All these theories are independent
but share an empiricist core.
Associative structures are usually contrasted with propositional structures, in
which individual elements are logically correlated. Propositional structures are
not just a causal sequence of associations. Associative structures express only
causal relations between representations, for example, the associative structure
green-tree says that we associate green with trees; it does not express the
proposition that (some) trees are green. The thought some trees are green
does not tell us that there is a causal relationship between the green-tree
representation. Instead, it predicates of some trees that are green.
Accordingly, associative inferences are transitions between thoughts that do
not follow from logical relations between elements of thoughts. In this sense,
associative inferences are contrasted with logical inferences, such as those
made use of in the computational-representational theory of thought, in which
inferences are truth-value-preserving transitions between thoughts that are
determined by their syntactic properties. Associative inferences are based not
on syntactic properties of thoughts but on associative relations between the
contents of thoughts. For example, we can associate the thought London and it
is raining because we once got wet in London. We cannot infer that it is raining
in London based on the formal properties of London and it is raining.
48 Thinking in Images

However, associationism does not appear to be a satisfactory account of

thinking in general. First, it is doubtful whether associationism can explain
the predicative nature of thinking. The thought trees are green predicates
something of trees. It does not just causally connect thoughts about trees and
colours. Associative structures cannot explain how thoughts can predicate
something of the world.
Second, the same thoughts can appear in different intensional contexts. I can
believe that-p and hope that-p. Associationism can explain it only if it interprets
the thought ‘that-p’ as different across various intensional contexts, which is
what we are trying to avoid.
Third, thoughts are compositional and systematic (e.g. Fodor and Pylyshyn,
1988). The associative structure of thought that binds together the list of
representations trees, are and green is compositionally indistinguishable
from the thought trees are green. However, these are two different structures.
Moreover, understanding the thought John loves Mary presupposes that we
also understand the thought Mary loves John. It is unclear how to explain this
systematicity in a model in which, for example, past experiences have not linked
Mary and John to a mutually loving relationship.
Fourth, some thoughts are coextensive. For instance, the thought that
Cicero is a Roman philosopher is coextensive with the thought that
Marcus Tullius is a Roman philosopher. Associationism captures the
difference between these thoughts. It is not able to explain what the identity of
these thoughts consists in.
In contrast, propositionalism explains rationality and the possibility of
modelling thinking according to logical rules. It explains the phenomenon
of systematicity and the compositionality of thoughts. It gives us an elegant
description of how thoughts can appear in different intensional contexts.
It provides an answer to the question of coextensionality and the predicative
nature of thoughts.
Yet the greatest strength of propositionalism is the weakness of the competing
theory, for accepting the propositional theory of thought does not come without
a cost. First, it forces us to make a number of metaphysically loaded assumptions,
such as that humans are rational beings or that rationalism best describes the
phenomenon of thinking. Consequently, we must explain the place of rationality
and thought in the physical world. Frege’s explanation leads to an acceptance
of Platonism; Fodor’s view leads to the endorsement of the idea of an innate
language of thought. Second, propositionalism cannot capture the phenomenon
of non-propositional thinking, such as imagistic thinking.
 What Is Thinking? 49

How does one find a place for imagistic thinking within this division? There
are two dominant strategies. First, we can try to interpret images in propositional
terms, trying to adjust images to the requirements set by propositional theory
(e.g. Blumson, 2012; Camp, 2007). Second, we can bite the bullet and claim
that imagistic thinking is more appropriate to associationist theories of thought
(e.g. Quilty-Dunn and Mandelbaum, 2020).9 However, both strategies are
unsatisfactory. On the one hand, propositionalism cannot provide a theory of
imagistic thinking, for images are non-propositional in nature. On the other,
adopting an associationist theory of imagistic thinking is running away from
the problem.
Granted, associationism is not a false theory. After all, thinking can take
different forms. When I think that trees are green, I can infer that some
trees are green, but I can also think of summer. However, one wants to choose
propositionalism over associationism because the first can provide a theory of
rational thinking while the second cannot.
We want to ensure that rationality may be expressible in imagistic thinking since
images are crucial for science and can work as arguments in reasoning. Moreover,
according to the Translatability Thesis, we should be able to transform images into
propositional forms and model relations between images according to the rules
of logic. We certainly do that. Operations on sets can be performed on both Venn
diagrams and propositions. Any theory of imagistic thinking should be able to
include inferential transitions between images. Therefore, it has to respond to the
challenges set by propositionalism and not be left at the mercy of associationism.
What is the lesson? It seems that neither propositionalism nor associationism
offers a theoretically respectable way of incorporating images into a coherent
theory of thought. Thus thinking with images cannot be explicated within the
propositionalism–associationism distinction. However, this problem is deeper.
It stems from the impossibility of freeing oneself from the traps of metaphors
that determine our thinking about thinking.

Metaphors of thinking

Although the distinction between propositional and associationist theories has

dominated philosophical discussions, it is often forgotten that it is underpinned
by a more basic understanding of the nature of thinking. It is expressed through
intuitions and metaphors rather than sets of theories. For this reason, it is rarely
directly tackled.
50 Thinking in Images

There are two dominant metaphors for what thinking is that are at the heart
of different theories of thinking (Schooler et al., 1995). According to the moving-
through-space metaphor, thinking involves moving through a logical or physical
space. Thinking is something we do and for which we can determine certain
methods and rules in the same way as we can determine pathways in space. We
use the spatial metaphor when we use such phrases as ‘searching one’s mind’,
‘approaching the problem from a different angle’ or ‘changing the direction of
thought’. We are also willing to answer the question ‘What are you doing?’ with
‘I am thinking’. According to the spatial metaphor, thinking is a kind of activity
at which we can be better or worse. We can be ‘too tired to think’ or we can be
‘deep’ or ‘slow thinkers’.
According to the perceptual metaphor, thinking is an act that happens to us,
often unconsciously and automatically. Thinking is a kind of seeing, not doing.
We use the perceptual metaphor when we use such phrases as ‘seeing a solution’,
‘gaining insight into the problem’ or ‘casting light on something’.
These two metaphors do not overlap with the distinction between propositional
and associationist theories of thinking. Frege, computational theories and
behaviourism adopt the spatial metaphor. The perceptual metaphor has been
developed within theories of divine illumination, the Cartesian theory of clear and
distinct ideas, Locke’s theory of ideas and Gestalt theory (Figure 2.1).10
The moving-through-space metaphor lies at computationalism’s heart (Fodor,
1975; Newell and Simon, 1972). Solving a problem can be seen as moving through
a problem space from an initial state to the goal state. Movement through the

Figure 2.1 Theories of thinking. © Piotr Kozak.

 What Is Thinking? 51

problem space requires the use of operators. These are actions that fulfil certain
subgoals. Moreover, holding propositional attitudes can be interpreted as
occupying a place in logical and physical space and as rule-governed transitions
from one state to another. For instance, the belief that grass is green can be
considered a state in a logical space isomorphic with the corresponding physical
exemplar of the sentence ‘grass is green’ at a particular location in the human brain.
This sentence may be the basis for further computational processes comparable
to movements through the state-space, such as the belief that my lawn is green,
which, in turn, can be the basis for action, for example, to water the lawn.
The same moving-through-space metaphor forms an underpinning of the
behaviourist theory of thinking. Behaviourists, such as Skinner (1957), held
that mental states could be interpreted as abbreviations of certain actions. For
instance, the feeling of sadness is an abbreviation of certain external behaviours,
such as crying and facial expressions, and internal behaviours, such as neural
activations, muscle tension and so on. In this vein, they assume that thinking
is a ‘silent speaking’, manifesting in subtle behaviours such as laryngeal muscle
movements.
However, the spatial metaphor does not seem to fit theories of imagistic
thinking. Entertaining an imagistic thought of a sunny day in summer does not
seem to be a movement in a logical space. Instead, it is a form of perception.
In particular, the theory of thinking as silent speaking does not seem to apply
to musical or visual thinking acts. When a painter thinks of a landscape, his
thinking process takes the form of perceiving an image rather than speaking to
himself.
The perceptual metaphor is storied. It has its roots in the Aristotelian theory
of intellectual intuition. It has been used, among others, in medieval theories
of divine illumination, the Cartesian clear and distinct ideas, Locke’s theory of
ideas and Gestalt Psychology.
However, the perceptual metaphor is at loggerhead with what we know
about perception and thinking. On the one hand, philosophical theories cannot
explicate the difference between perception and thinking (more on that in
Chapter 3). On the other, even though it is promising, Gestalt Psychology is far
from providing a comprehensive theory of thought.
Gestaltists analyse thinking in problem-solving contexts and emphasize
insight’s role in understanding a problem’s structure. Within Gestalt Psychology,
understanding is not an incremental and continuous process but is sudden and
spontaneous, similar to an act of perception. However, Gestalt Psychology is
fairly accused of being uninformative (e.g. Jäkel et al., 2016). It does not explain
52 Thinking in Images

phenomena but describes them. It does not indicate the mechanisms governing
the thinking process but describes the qualitative structure of the thinking
process. It does not explain, for example, what a sudden process of insight
consists in. Thus, it cannot be a candidate for a theory of imagistic thinking.
Where do we stand? On the one hand, the spatial metaphor is inconsistent
with how we use images in thinking. On the other, the perceptual metaphor
does not explain what a theory of imagistic thinking could be. However, it is
forgotten that these two metaphors are not mutually exclusive. Although they
are frequently discussed independently, they are complementary.11 When
moving through a physical space, we have to see where to go next. In problem-
solving, movement from an initial state to the goal state requires recognition of
the correct operators. Thus, these two metaphors can be readily combined. With
respect to thinking, such a combination can be beneficial because it suggests that
multiple processes may contribute to what we call thinking. It also suggests that
numerous factors may contribute to an impasse in thinking. On the one hand,
one can know how to move through a logical space but fail to see the goal. On
the other, one can see where to go but not know how.
A more adequate metaphor for thinking would be that of measurement.
Thinking is like applying measures, and thoughts are the products of those
applications. The reasoning is to see what results from applying different measures.
The measurement metaphor contains the perceptual and spatial metaphor.
Measurement is the result of a certain action. For instance, it may be the action
of measuring the time between thunder and a flash of lightning. It is moving
through logical space by determining spatial and temporal relations between
measured objects. Yet, to conduct a measurement, one has to recognize what is
to be measured – in this case, distance from the storm. With this metaphor in
hand, we can consider alternative ways of building theories of imagistic thinking.

Summary

The Received View is a loose collection of beliefs and intuitions about the nature
of thinking, according to which thinking is propositional and action-based.
Thinking is a matter of operations carried out on propositional structures made
of concepts.
The Received View gives a clear idea of what knowledge and the relation
between knowledge and thinking are. It provides a theory of the content of
thoughts. It is in line with metaphysical assumptions concerning the nature of
 What Is Thinking? 53

the bearers of thought. Thoughts, as well as propositions, are compositional and

systematic.
Imagistic representations do not fit into the Received View. They seem to be
unable to provide a theory of knowledge and a theory of content. They seem
to fail to explain the systematic and compositional nature of thoughts. Thus,
if we accept the Received View, the concept of thinking with images is self-
contradictory.
Any theory of imagistic thinking requires broadening the notion of thinking.
At the same time, it should be able to address the epistemological, semantical
and metaphysical challenges directed against such a theory.
54
 3

What answers should we expect?

Bertrand Russell noted (1919, 11) that ‘If you try to persuade an ordinary
uneducated person that she cannot call up a visual picture of a friend sitting
in a chair, but can only use words describing what such an occurrence would
be like, she will conclude that you are mad’. However, the claim that we think
with images has always been subject to suspicion, for our prescientific intuitions,
which supposedly support imagistic theories of thought, have always been vague
and fallible. That is a simple consequence of our lacking a clear understanding
of what these intuitions are thought to support. The imagistic account of human
thought is a source of misunderstandings and disputes mostly because it is not
clear what we really mean when we say that we think with images. As Pylyshyn
(2003b, 113) observed, despite claims that thoughts have a picture-like format
having persisted for such a long time, the problem of stating clearly what it
means for thought to be imagistic has rarely been explicitly tackled. Yet the real
difficulty lies deeper: we do not even know what kind of answers we expect when
we ask, ‘What is thinking with images?’ To know the content of a question is to
know what counts as valid and invalid answers to the question. To know the
content of the question ‘What is thinking with images?’ is to know what kind of
answers we should expect. In other words, we should know what counts as an
explanation of what thinking with images is.
We have to be careful not to trivialize the answer. We face the problem of
triviality when we try to describe a phenomenon without explaining it. The
problem of triviality stems from the intuitive character of imagistic theories
of thought. At first glance, it seems like there is nothing wrong with imagistic
thoughts and that they do not need any clarification: the claim that we can
think with images sounds intuitive, and we have to accept it as a brute fact.
This prescientific intuitiveness is the cause of our continued attachment to the
imagistic theory of thought, but it is also a significant source of conceptual
confusion.
56 Thinking in Images

This commonsensical intuition found its early scientific expression in

Francis Galton’s (1880) famous questionnaire On Visualizing and Other Allied
Activities, which analyses, among other things, different styles of thinking. It
would go on to receive further scientific attention in a number of studies on the
nature of memory in the early 1960s (Baddeley and Lieberman, 1980; Baddeley,
1988; Paivio, 1963; Shah and Miyake, 1996) and research conducted in the
chronometry paradigm (Shepard and Metzler, 1971), building what has been
commonly called ‘the pictorial approach’ to the nature of mental representations
(e.g. Kosslyn et al., 1994, 2006; Pearson et al., 2015).
Considerable worries arise when one tries to accommodate the trivial fact
that images play an important role in cognitive practice with philosophical
accounts of the nature of thinking. For from the fact that one commonly uses
images, it does not follow that one thinks with images, just as the observation
that we frequently use commas in writing says nothing about what language is.
Therefore, without any supplementation, the intuition that we think with
images is either trivial or false. Either it says only that we commonly use (mental)
images in cognition, which nobody disagrees with, or that from the frequent use
of images in thinking we can directly infer the true nature of thinking, which is
false. Without establishing what kind of contribution images make to thinking,
nothing follows for the explanation of the latter. We need an argument as to why
we consider using images an instantiation of thinking and not something that
accompanies thinking.1
There are two dominant non-trivial strategies for answering the question
‘What is thinking with images?’, depending on how the notion of the bearer
of thought is understood. According to the neo-Lockean approach, imagistic
thoughts are theoretical objects that mediate between perceptual and discursive
representations. According to Wittgenstein–Ryle’s approach, the question is
senseless, for there are no objects we think with.
In this section, I argue that both strategies fail to address the problem of what
thinking with images is. I introduce an alternative strategy for answering this
question that is put in terms of the operational approach. According to it, the
imagistic bearer of thought can be understood in terms of the family of operations.

Imagistic thoughts as theoretical objects

The leading cause of the revival of imagistic theories of thought in the twenty-
first century is a belief that a full-blooded theory of thought should explain
 What Answers Should We Expect? 57

how we acquire thoughts. According to the explanatory strategy represented by

Locke2 and neo-Lockean approaches, such as Barsalou’s (1999), Gauker’s (2011)
and Prinz’s (2002), images offer a plausible answer to the problem of thought
acquisition. The basic idea is that we have no choice but to accept the imagistic
theory of thought to solve empirical and philosophical problems concerning the
nexus between perception and thoughts. According to this strategy, we posit
the existence of some mental objects, such as imagistic thoughts and Lockean
ideas, just as we hold in astrophysics that to solve problems with the accelerating
expansion of the universe, we postulate the existence of dark energy. In both
cases, we claim that images and dark energy are necessary for explaining
certain psychological and physical phenomena. We postulate the existence of
some physical and mental objects to make some phenomena explainable. The
explanation of what imagistic thoughts and dark energy are would be similar:
imagistic thoughts and dark energy are theoretical concepts that we adopt to
explain certain phenomena. Thus, we need to know what problems images solve
and how.
As Locke states in the opening chapters of his Essay Concerning Human
Understanding (1975, 104–5):

Let us suppose then the mind to be, as we say, white paper, void of all characters,
without any ideas: How comes it to be furnished? [. . .] Whence has it all the
materials of reason and knowledge? [. . .] To this I answer, in one word, from
experience. [. . .] First, our Senses, conversant about particular sensible objects,
do convey into the mind several distinct perceptions of things, according to
those various ways wherein those objects do affect them: and thus we come by
those ideas we have of yellow, white, heat, cold, soft, hard, bitter, sweet.

To understand the Lockean innateness-free strategy of argumentation, one

must bear in mind that Locke’s ontological argument is a consequence of an
epistemological one. Ontological features of the mind are determined by analysis
of cognition. The rejection of nativism is based on a view of how cognition
and knowledge are to be possible. Introducing the imagistic theory of thought
follows from the fact that it helps us find a direct link between thoughts and
perception, grounding our thoughts and cognition in perceptual experience,
thanks to which, in short, we can keep our thoughts in touch with the empirical
world.3
Why is addressing the problem of grounding thoughts in perception
epistemologically important? Linking perception and thoughts seems to provide
a plausible naturalistic explanation of the genesis of our thoughts. It helps to clarify
58 Thinking in Images

the ontogenesis and phylogenesis of our thoughts. Hence, linking perception

and thoughts should provide a plausible description of how our thoughts stem
from perceptual experience and what constitutes this possibility, that is, what
kind of conditions one would have to accept if this link were possible. Burge
(2010) and Prinz (2002) are, among others, prominent representatives of such
an approach.
The great advantage of the imagistic theory of thought is that it is relatively
straightforward to imagine how thoughts could be inherently perceptual,
deriving directly from experience. From a general perspective, there is relatively
little difference in how we characterize this link, whether it is based on the
abstraction of simple sensations to more abstract ideas (e.g. Hume, 1977; Locke,
1975) or on a common perceptual symbol system (Barsalou, 1999). What is
essential is that the imagistic model of thought makes this link conceptually
possible, or, to put it differently, it would be more difficult to understand how
any non-perceptual model of thought could explain it.
The problem of binding thoughts and perception is based on how one
reconciles the categorically different features of two different kinds of mental
acts. It is believed that perceptual representations have iconic content and
format while thoughts are discursive. The high hopes put on the imagistic
theory are related to the fact that it introduces a link between thoughts and
perception. In the most general sense, the role of an image is to mediate
between perception and thinking, being both the product of perception and
the basis for thoughts at the same time. Playing a mediatory role means here,
among other things, that images share features that can be ascribed to both
perceptions and thoughts.
The mediatory role of images is usually understood as the claim that it is
very likely that conceptual tasks, such as concept acquisition, categorization
and inference, rely on perceptual representations (e.g. Goldstone and Barsalou,
1998). It is supported empirically by the body of evidence produced mainly in
the 1990s and 2000s (e.g. Kan et al., 2003; Kiefer and Pulvermüller, 2012; Pecher
et al., 2004).4
Without going into details, these results suggest that there is a link, presumably
iconic, between thoughts and perception. It builds what Machery (2016) calls
the ‘neo-empiricist consensus’, which united philosophers, psychologists and
neuroscientists at the beginning of the twenty-first century (e.g. Barsalou, 2010;
Martin, 2007; Prinz, 2002). A significant advantage of the imagistic theory of
thought seems to be that it provides a reasonable picture of the nature of the link.
Images seem to mediate between thoughts and perceptions, sharing features
 What Answers Should We Expect? 59

of both; thoughts are grounded in perception in virtue of being mediated by

imagistic representation.

Thoughts–perception border

Although it seems as if the claim that thoughts are perceptually grounded

provides a convincing justification for the imagistic theory of thought, it appears
to be misleading upon closer inspection. For one thing, it blurs the distinction
between thoughts and perception. For another, it appears that we overestimate
the credibility of the case that seems to support the claim.
First, perceptions are not reducible to thoughts. Early perception is believed
to be mostly5 cognitively impenetrable and possesses a different kind of content
than thoughts (Firestone and Scholl, 2016; Pylyshyn, 2003a; Raftopoulos, 2009;
Raftopoulos and Zeimbekis, 2015). Moreover, the claim that thoughts and
perceptions share a common representation format is insufficiently supported
by empirical evidence. A vast number of studies show that thoughts, in contrast
to perceptions, are amodal.6
Second, the idea that the content of a thought is perceptually grounded
may be understood as the claim that every concept has a direct counterpart in
an associated image that grounds the concept's meaning. The idea, however,
faces obvious objections. Many concepts do not stand for any kind of image;
for example, we do not need images for concepts like and, not, yesterday,
tomorrow, inflation, prime, infinity or negation to understand their
meaning. It is even difficult to imagine how we could have an image for empirical
concepts such as photon. And even if we had such images, it would be hard to
determine what they mean. Mental images are private and cannot be the basis
of public concepts. Thus, the Lockean account of thinking cannot provide a
comprehensive theory of thought and was rightly rejected in the middle of the
eighteenth century.
And here is the dilemma: although we have strong reasons to believe that
thoughts are perceptually grounded, we have to admit that we have equally
strong reasons to believe that we cannot interpret the nature of thought in purely
perceptual terms. Perception and thinking are distinct mental capacities. It may
be true that the border between thoughts and perception is not easy to determine,
and there are many controversies about where the border is (e.g. Clark, 2013).
Yet it does not follow that there is no border at all (e.g. Green, 2020). If all of
that holds, then it may be doubted whether images may be common ground
60 Thinking in Images

for perception and thoughts since that very commonality is logically excluded.
Images cannot put perception and thought together like some other kind of
representation cannot put colours and numbers together.
The difficult task at hand is to understand how to reconcile the claim that
thoughts are perceptually grounded with the thesis that there is a border
between thoughts and perception. Imagistic thoughts are believed to cover both
cases. Therefore, they are supposed to possess characteristics that would enable
us to interpret them as perceptual but not as instantiations of the genera of
perception. Thus, any neo-Lockean explanation of what thinking with images is
would have to be general enough to spell out what connects imagistic thoughts
with perception and specific enough to point out what differentiates them. To
make it clear, it is not logically excluded that it can be done.
Moreover, it is the case that every theory of thinking has to face this dilemma.
The point is that it cannot be done simply by invoking a third term – an image –
that mediates between perception and thoughts. Let me illustrate this problem
with two contrasting neo-Lockean approaches: Jesse Prinz’s (2002) theory
of perceptual concepts and Christopher Gauker’s (2011) theory of imagistic
thinking.
The Lockean way of thinking about the perceptual grounding of thoughts
was based on the idea that thoughts are abstract objects with an image-like
nature. It may mean that the meaning of mental representations is grounded
in relation to a corresponding mental image. However, such an interpretation
is nonsensical, and it seems to have the same scientific status as the geocentric
theory in astrophysics.
A more sophisticated way to spell out the neo-Lockean theory is based on
the idea that mental representations do not require an imagistic counterpart to
ground their meanings but have an image-like nature. Let me illustrate this
concept with Prinz’s proxytype theory.
According to Prinz (2002, 149), proxytypes are ‘mental representations of
categories that are or can be activated in working memory’. The central idea
is that thinking is a simulation of perception – it is an idea borrowed from
Larry Barsalou’s seminal paper on perceptual symbol systems (1999) – and to
think about something is to put oneself into a state that resembles the state of
perceiving it. For example, the concept dog is a set of perceptual representations
acquired when one encounters, imagines or hears stories about dogs. All the
experiences of dogs one has met, all of the images of dogs one has seen, all
the stories one has heard, plus mental links that bind together those states build
the concept dog. On a given occasion, specific subsets of the set can be recruited
 What Answers Should We Expect? 61

to recognize or classify items in the world and form parts of thoughts as parts of
larger imagistic or perceptual scenarios – these are proxytypes. They are context-
sensitive. For example, if I think about a watchdog, I will invoke a representation
of a German Shepard Dog and not a Poodle, but if I think about a cute small
dog, I will invoke an image of a Poodle and not a German Shepard Dog. The
‘proxytype’ we construct on a given occasion depends on our cognitive needs in
the moment.
The main advantage of the proxytype theory is that it helps us make sense of
concept empiricism – the idea that all concepts are grounded in perception. It
also explains some phenomena in cognitive psychology. For example, it explains
the retrospective nature of representations that underlie many of the violation-
of-expectance looking-time experiments or lags in understanding that manifest
themselves in infant looking-time studies and tasks requiring explicit imagistic
representations (e.g. Carey, 2009).
However, it is doubtful whether the theory of proxytypes solves the binding
thoughts and perception problem, for the theory does not meet the Semantical
Challenge. Notice that when one invokes two perceptual representations – a
representation of a German Shepard Dog and a Poodle – we have two different
representations. They do not represent what they have in common. The concept
dog does that. If concepts were perceptual representations of some general
kind, such as a perceptual representation of a canine, we would need another
perceptual representation that would determine the similarity between the
representation of a canine, the German Shepard Dog, and the Poodle, and it
would not stop there. We would need another representation, ad infinitum.
Moreover, a concept like the concept dog has a place in propositional
structure. By substituting a representation of an object into an argument,
we can form a proposition, such as all dogs are mammals. If the concept
dog were a perceptual representation of the same kind as other perceptual
representations, such as a representation of a mammal, there would be no way to
determine whether the representation dog plays the role of a named argument
or a predicate in a proposition. In other words, if mammals and dog were both
perceptual representations, it would be impossible to distinguish between two
different propositions like All dogs are mammals and All mammals are
dogs. Prinz recognizes the problem and tries to develop the idea of mental links
that bind together perceptual representations and determine the functional roles
of perceptual representations. But what he achieves is introducing concepts
under a different name, implicitly restating that perceptual representations alone
cannot be concepts.
62 Thinking in Images

Gauker (2011, 2017) approaches perception and concepts differently. Prinz

posits that there is a continuity between thoughts and perception because
concepts are perception-like. However, it contradicts the claim that there is a
border between perception and thoughts. Gauker’s idea is that there is continuity
and a border between thoughts and perception, for thinking can be divided into
two kinds. The first kind of thinking is conceptual and language-dependent. It
introduces the border between perception and thoughts. The second kind of
thinking is non-conceptual and imagistic – it introduces the continuity between
perception and thoughts.
The general idea is that perceptual representations cannot be the basis of
concepts and abstract representations. Perception of a table is always a perception
of a particular object and cannot represent kinds such as a piece of wooden
furniture or table – language can. Gauker’s alternative is to demonstrate that
thinking can consist in non-conceptual operations based on mental images. To
explain the nature of imagistic thinking, Gauker holds that we can see imagistic
thoughts as regions of perceptual similarity spaces and as topological maps of
objects in the brain.
Imagistic thinking is described as a capacity to locate and track objects
and scenarios in perceptual similarity space. A perceptual similarity space is
a topological hyperspace in which the dimensions represent various ways in
which objects may perceptually differ. A mark in a perceptual similarity space is
a vector of the perceptual similarity space. The location of a mark on a dimension
of the perceptual similarity space is the mind’s measure of the location of a
sensory object or a scenario on the dimension of the objective quality space
that corresponds to it. For example, when the mark representing the German
Shepard Dog is closer in perceptual similarity space to the mark representing
the Poodle than a Persian cat, then a subject assesses that the German Shepard
Dog is more similar to the Poodle than to a Persian cat. The general idea is that
similarity spaces can explain how we recognize and group objects in perception
without postulating language-driven concepts.
Gauker’s idea is that we can think about thinking not in terms of operations
on concepts but in terms of perceptual similarity spaces and imagistic cognition.
According to Gauker, concepts are a matter of mastering language. Thus,
concepts are building blocks of language-like propositions. However, not all
thoughts are conceptual. The medium of conceptual thought is language. The
medium of non-conceptual thought is an image (Gauker, 2007). To represent
an object imagistically is to assign it to a location in the perceptual similarity
space.
 What Answers Should We Expect? 63

However, there are two problems with Gauker’s non-conceptual view of

imagistic cognition. First, the problem concerns a transfer between conceptual
and non-conceptual representations. Imagistic thoughts cannot be easily put
into words, but that does not mean that images cannot be put into words at
all. For instance, I can imagine a cat sitting on a mat, expressing my ability to
recognize objects in perceptual similarity space, but I can also form a definite
description ‘the cat sitting on a mat’.
The problem is that we want to maintain the possibility of translation between
perceptual and conceptual thoughts. However, such a possibility is excluded since
they share different metaphysical properties. According to the Metaphysical
Challenge, conceptual thoughts are compositional and systematic; imagistic
thoughts are not. Translating a non-conceptual system of representation into a
conceptual one is not like translating German into English but like trying to put
a set of randomly generated phonemes into English.
Second, one of the main advantages of Gauker’s theory of non-conceptual
imagistic thought is its capacity to explain the behaviour of non-linguistic
animals, such as non-human animals and infants. Imagistic thinking seems to
be more primitive and more fundamental than language and language-based
knowledge. However, the problem arises when we try to extend the thesis of
the non-conceptual content of imagistic thinking onto the domain of scientific
knowledge. Pictures can be a source of knowledge, as scientific practice
shows. We can ask, however, what is the nature of such knowledge. It could be
tempting (e.g. Mößner, 2018) to argue that pictures and visual perception are
bearers of non-conceptual knowledge and can show something that is not yet
conceptualized. So far, we seem to align with Gauker’s theory of non-conceptual
content. Yet, as the Epistemological Challenge shows, it cannot be the case.
Science consists of theories and hypotheses that are built from propositions and
concepts. The goal of science is to put things in the conceptual form of theories
and hypotheses. Any content that has not yet been conceptualized cannot be
part of scientific knowledge.
Thus, it seems that neo-Lockean strategies fail to explain the thought-
perception problem. The proxytype theory does not meet the Semantical
Challenge. Gauker’s theory of imagistic cognition does not meet the
Epistemological and Metaphysical Challenges. It does not mean that the
thought-perception problem has disappeared. Even though we do not have a
satisfactory solution – and because of that – we still need one. The point is that
the problem cannot be easily solved by invoking imagistic representations as
theoretical concepts that can make this problem solvable. The problem is even
64 Thinking in Images

more profound. It is controversial whether postulating the existence of images as

mental objects we think with makes any sense.

Wittgenstein–Ryle’s sceptical argument

The main philosophical problem with the question ‘What is thinking with
images?’ is connected to the overwhelming temptation of explaining away the
problem, which is based on the assumption that the very question is ill-posed.
As one may argue, it is not the case that we think with images, just as it is not the
case that we think with or in words.
This claim is most frequently expressed by the view that we think neither in
images nor in words but in thoughts. Although such an answer to the question
may seem tempting, it is nonsensical. To say that we think in thoughts is no
more informative than saying that we do not speak in French or English but
use language. The misleading character of the answer follows from the fact that
to study language, we study particular instantiations of language – for instance,
French or English. The same is true in the case of words and images. We can only
study the nature of thoughts by studying particular instantiations of thought
processes expressed in language and images.
A more sophisticated way to render the question absurd is to argue that the
very expression ‘thinking with’ is senseless. Expressions such as ‘thinking with
words’ or ‘thinking with images’ lack meaning. Wittgenstein (1953, 1958) and
Ryle (1979) undertook this sceptical line of argument (Bennett and Hacker,
2003; Slezak, 2002b).
Let me illustrate this strategy with Ryle’s (inspired by Wittgenstein) argument
against the belief that thoughts consist of word-like objects that are expressed
in language. According to Ryle (1979), it would be senseless to say that one
composes their speech expressing some English-like thoughts. Considering
English words and phrases, modifying and rejecting others, stringing together
words into sentences and paragraphs, one is not thinking in some English-like
medium. Their speech will be, of course, in English, but we cannot say that the
mental process of composing the speech was in these words. They did not have
these words while composing the speech. They were searching for these words
in the process of speech composition. Thus, the expression ‘thinking with words’
is senseless.
The general objective of a Wittgenstein–Ryle style argument is to undermine
a dominant view on the nature of thinking, which could be put in terms of a
 What Answers Should We Expect? 65

vehicle-cargo model. According to the model, thoughts consist of the content of

the thought – something the thought is about – and the bearer of the thought –
something the thinking takes place in, for example, in words, in images or any
other representational system. According to Wittgenstein and Ryle, this model
leads to nonsensical consequences and should be abandoned.
As clearly stated by Bennett and Hacker (2003), the source of the confusion is
the idea that we think in a medium, just as we speak in a language. We can even
say that we think in English, and not in German, and if not in English, then in
French. And even if we do not think in a language, we must think in something
else, for example, in images. But that is only a conceptual confusion.
Talking to oneself is neither necessary nor sufficient for thinking. I can talk
to myself without thinking, for example, when counting sheep, to stop thinking
and fall asleep. I can think without talking; for example, I can directly formulate
a conclusion based on some evidence, without saying that p follows from q, if r,
and so on. It does not mean that thoughts are primarily in images, being a basis
for translation from images into words. To use a word to express a thought has
nothing to do with having an associated mental image. According to the private-
language argument, if having an associated mental image were necessary for
understanding the meaning of a word, then it would be impossible to determine
the meaning of the word. Thus, one need not think in anything; particularly, one
need not think in images.
Moreover, the sceptical line of argument could even be strengthened by
pointing out that whether we think in language or images is misleading in two
different ways. One can argue that we do not have to think in language but must
possess the language to express thoughts. As an example, we need language to
express a rational train of thought. Note that there are only thoughts that could
be (but do not need to be) expressed. The concept of an inexpressible thought is
in the same sense senseless as a function that does nothing. Therefore, we have
to master the language to be able to think rationally. Thus, the words one uses
when one thinks are the expressions of one’s thought.
In contrast, images are neither necessary nor sufficient for expressing a
rational train of thought. Images may cross one’s mind while reasoning, but
they can only play the role of heuristic devices to help one find a solution. They
can aid thought as an accompaniment of thinking, but they are not thoughts or
expressions of thoughts.
However, we should not feel intimidated by these arguments. Following
Goodman (1984), there is a notable linguistic difference between words, images
and other objects we can think of. Linguistically speaking, it seems acceptable
66 Thinking in Images

to say that we can think in words and in images, as well as of words and
images. In contrast, we can think of cabbages, but we cannot think in cabbages.
Analogously, we can speak in words without speaking of words, and we can
speak of cabbages but not in cabbages. What does it mean to think in words or
in images?
First of all, it cannot be reduced to thinking of words or images, just as worship
with images is not reducible to the worship of images. Thinking of cabbage is not
reducible to thinking of a word or an image representing a cabbage since the
thought of a word or an image would need another word or image, and so on,
ad infinitum – that was Ryle’s primary point. Words and images are somehow in
mind, while the cabbages, words and images we think of are not.
Second of all, thinking of words and images and thinking of cabbages do
not require producing them. The phrases ‘thinking in words’ and ‘thinking in
images’ do not mean that thinking is a three-term operation between thinking,
the object of thought and some entity, a word or an image, that moderates
between the act and the object of thought.
In contrast to the Lockean approach, images do not have to be postulated
mental objects. I can think of a hexagon and draw a hexagon without forming an
image of a hexagon in my mind. Such a proposal does not require presupposing
objects such as words and images we think of to substantiate thinking in words
and images. Instead, thinking like speaking and manipulating images may be
interpreted, in Kantian and Fregean fashion, as an operation conducted in the
mind, where mental and extramental images may be interpreted as the final
results of those operations and necessary means to represent these operations.
The term ‘operation’ is polysemic. In general, it means a rule-governed process
of taking input values into the output values. Notably, operations are not limited
to computations. Instantiations of operations are subtraction, construction,
functions mapping propositional contents onto a set of truth values, a partial
ordering function and so on. Two points have to be stressed. First, the
distinction between operations and the results of operations is not the same as
the distinction between actions and the products of actions (Twardowski, 1999).
To be an operation is to be a rule-governed process, for example, a function,
regardless of whether the operation is carried out in a relevant action.
Second, operations and the results of operations are different sides of the same
coin. Every operation can be reformulated into a relevant result of an operation,
and vice versa. For example, mathematical axioms can be reformulated into
inference rules and the other way round. From a logical point of view, their
status is equivalent (e.g. Stegmüller, 1969).
 What Answers Should We Expect? 67

Therefore, we have to distinguish between images as results of operations and

image-producing operations. The claim that there is no image in mind mediating
the process of thinking in images does not mean that there is no operation that
produces an image. Consequently, when we inquire into the nature of thinking
in images, we are not inquiring into the nature of objects such as images in mind
but the nature of image-producing operations, resulting in relevant images. In
other words, our questions are about the nature of a family of operations labelled
‘thinking with images’.

The operational approach

Interpreting the relation ‘thinking with’ in terms of operations helps us better

understand the nature of the concept of a ‘bearer of thought’. In short, the notion
of a bearer of thought is interpreted here in terms of a family of operations
conducted in a representable medium. The basic idea is that thoughts are
operations that need certain vehicles to be carried out and represented, such as
words or images. In the same way, counting is a function that to be carried out
requires that we introduce certain vehicles, that is, numbers we count with and
numerals that make those numbers representable.
Importantly, numbers do not exist outside of those operations, for example,
outside of the context of counting. If one asks what numbers are, one does not
ask about some objects in Platonic heaven. One asks what numbers do, that
is, what kind of operations they can be part of. For example, if one asks what
numbers are, a valid answer may be ‘something we count with to determine a
less-than relation between a and b’. At the same time, numbers can be seen as the
final results of these operations. For example, one can say that the number 2 is
the final result of an operation determining the cardinality of a set; that is, it is a
measure of the set’s size. To be the number 2 is to be more than 1, 0 and less than
3, 4 and so forth. Searching for numbers as ideal Platonic objects is misguided.
Moreover, numbers require numerals to be represented, for the limits of
possible operations are the limits of possible representations of the operation.
It means that every number ought to be representable, though not necessarily
actually represented, by a relevant numeral. An unrepresentable operation
does not exist, for the nature of an operation is that it does something. It can
do something only in some representable medium, such as words, images and
numerals. An operation that does nothing is not an invalid operation. It is not
an operation at all.
68 Thinking in Images

The point is that just as there is no counting without numbers and without
numerals that represent them, images can be interpreted as necessary vehicles of
thoughts. In the same sense, some operations would be impossible to carry out
if no images instantiated and represented the operations. The nature of counting
and thinking with images is determined by the nature of the operations, not by
the nature of the vehicle, such as the features of numerals and pictures. Numbers
can be represented by digits, words, such as ‘one’ and ‘two’, and apples. What
determines that an expression is a numeral is not that it consists of words or
digits but the fact of what the expression represents. Every representation of a
number is a numeral, whether formed in words or apples. To understand the
nature of numbers, one must understand what kind of operation it is a part of.
For example, to understand the number 2, one must understand that 2 + 2 = 4 or
that a set contains two and only two elements.
By the same token, to determine what it is to count and what it is to think
with images is to determine what we do, not what numerals or pictures hanging
on a wall are. At the same time, numbers and images are not tools to think with,
where the nature of the operation of using a tool is separable from the features
of the tool. Numbers are not like calculators, for it is not the case that when
learning how to count, one first learns how to perform calculations and then
what numbers are. Numbers are necessary if operations like counting are to
be carried out. Similarly, images are necessary for thinking in the sense that
some operations would not be possible without them. For instance, it would be
nonsense to say that we first learn how to form an image, such as a drawing, then
try to find tools to draw. Learning how to draw is inseparable from drawing.
According to the operational approach, imagistic thoughts are like numbers.
They are defined by the operations they are a necessary part of. Just like numbers,
imagistic thoughts require a representable medium, that is, images, to express
these operations. Consequently, the question ‘What is thinking with images?’ is
interpreted as equivalent to the question ‘What kind of operation is expressed
by images?’ The question may be confusing, just as it may be confusing to ask
‘What kind of operation is expressed by numerals?’, but that does not mean that
it is senseless. And even if the operations we are looking for are not the same
operations as the ones expressed in language, no one has to assume that they
have to be the same. Not all thoughts can be expressed in language; some can
only be expressed in images. The point is to find out what these thoughts are.
In contrast to the neo-Lockean approach, the operational view is non-
committal on the point of any mental objects and episodes displayed before the
mind’s eye, such as words or images we think of. The intuitive idea is that when
 What Answers Should We Expect? 69

we think of, for instance, what the result of an operation like 2 + 2 is, we do
not have to write down a formula on a piece of paper or construct an explicit
mental representation consisting of the symbols ‘2’ and ‘+’. We think in numbers
without having to think of them. However, any operation has to be conducted in
a representable medium, for example, in words and images.
Thinking about images in terms of operations may seem counter-intuitive.
A comparison to the propositional theory of meaning in the philosophy of
language offers a better picture of the point I am presenting.
According to the propositional view in the philosophy of language, the
meaning of the sentence S is the proposition P expressed by S. Much ink has
been spilled to argue that the propositional view is incomprehensible. For
example, we know neither how propositions exist nor whether they exist in
time and space. As a result, it could be argued in a Quinean (e.g. Quine, 1960)
fashion that we have to reject the propositional view as postulating ontologically
unnecessary entities.
However, the propositional view is non-committal on the point of the
existence of abstract entities such as propositions. For instance, intensional
semantics interprets propositions in terms of functions from possible worlds
to truth values. It means that propositions are operations that determine the
conditions according to which a sentence S is true or false in a possible world.
The indispensable role of S is to make this function representable. Therefore,
we can accept propositions as bearers of meaning without positing unnecessary
abstract objects.
The main point is that we can accept images and words similarly as bearers
of thoughts without being committed to the existence of image- or word-like
objects in mind. Images and words express some meaningful operations and
are indispensable for those operations to exist. The concept of a bearer of
thought is no more mysterious than the concept of a proposition, and therefore
Wittgenstein–Ryle’s argument misses the mark. The question of the nature and
function of imagistic representations in thought processes is the question of the
nature and function of bearers of thought in the same way as the question of
the nature of the meaning of words can be interpreted in terms of the question
of the nature of propositions. And although that does not mean that the question
of the nature of propositions is easy, it is not senseless.
One caveat is in order, the operational view is not reducible to the dispositional
view, and thinking with images is not reducible to a disposition to produce a
mental image (Price, 1953), to a certain skill (Bennett and Hacker, 2003; Ryle,
1949) or the state of readiness for producing images and for judging presented
70 Thinking in Images

images as agreeing with one in mind (Goodman, 1984). To put it more precisely,
interpreting the nature of an imagistic bearer of thought in terms of a disposition
does not suffice to explain the operations of using images in thinking.
Granted, possessing a disposition to produce and interpret an image is
necessary for thinking with images. In the same way, to understand the meaning
of a knight in chess is to have a disposition or a skill – interpreted loosely as a
regular disposition – to move it in a certain way.
However, it is not a sufficient condition. One can be disposed to think and
speak like Kant without any understanding of Kantian thought. One can have
the disposition to recognize similar pictures of a triangle without understanding
that the pictures represent a triangle. To master an operation, one has to know
what the correctness conditions of the operation are. In other words, one has to
know what counts as a correct instantiation of an operation.
In the same way, mastering the counting operation is to master a certain skill
concerning counting, but explaining the nature of counting is not the same as
explaining the nature of the corresponding disposition. Explaining the nature of
counting is to explain the rules governing a family of functions. An explanation,
such as a psychological explanation of the nature of the disposition, may be
part of an explanation of a function, yet it is not reducible to the latter.
To sum up, to make sense of the question ‘What is thinking with images?’,
it shall be interpreted here as a question concerning the nature of operations
expressed by images. It shall not be interpreted as a question concerning the
nature of some mental objects, such as imagistic thoughts, picture-like ideas
or mental images, or a question about the nature of dispositions. In short, the
images we think with are primarily something we do and not something we
possess.
Consequently, we can infer the nature of the given operation from studying
external instantiations of the operation. In the same way, we investigate the
nature of abstract objects, such as functions, by examining their instantiations,
for example, equations, and the nature of mental phenomena, such as linguistic
meaning, by studying linguistic representations.

Explaining operations

As I have argued, any attempt to answer the question ‘What is thinking with
images?’ has to begin with careful isolation of what is to be explained and an
understanding of what kind of answers may be considered satisfactory. For in
 What Answers Should We Expect? 71

the case of such operations as thinking, the very concept of explanation may lead
to confusion. The issue is particularly prominent in explanations of the nature of
such objects as gold and water.
When we try to explain the nature of gold, what is to be explained is the
nature of an object – a sort of yellow metal – and the explanans is the structure
of gold and the relations of gold to other elements. Let us say that gold is the
chemical element with the atomic number 79. We communicate the number of
protons in the element and the element’s position in the periodic table. If we say
that water is H2O, we are trying to explain the nature of some wet substance. Our
explanation is a description of the chemical structure of water and the relation of
water to other elements, for example, to oxygen.
In the case of thinking, the situation is different, for there is no thinking-like
object that is to be explained. As Ryle noted (2009, 271):

When we start to theorise about thinking, we naturally hanker to follow

the chemist’s example, namely, to say what thinking consists of and how the
ingredients of which it consists are combined. Processes like perspiring,
digesting, counting and apple-picking can be broken down into ingredient
processes, coordinated in certain ways. So it seems reasonable to expect thinking
to yield to the same treatment. But this is a mistake. There is no general answer
to the question ‘What does thinking consist of?’

In contrast to the object-like view, we can think about thinking as an operation, in

the same way as counting can be described in terms of a mathematical operation,
meaning in terms of mapping onto a set of possible worlds and language as a
set of semantic and syntactic rules. The result of thinking is a thought, in the
same way as the result of using a language might be a meaningful sentence, and
the outcome of a calculation might be a formula written on a piece of paper.
However, if we want to explain what thinking, language and calculation are, the
object of explanation is not some thought-, language- or counting-like object.
Suppose we are to understand what thinking, meaning, language and calculation
are. In that case, we must understand the nature of the operations expressed by
certain utterances, sentences and formulas. Therefore, explaining the nature of
thinking is explaining the nature of an operation. What does it mean to explain
an operation?
Notice that if one wants to explain the nature of an operation, such
as calculation, what is to be explained is not describable in terms of the
phenomenology of calculation. In that context, phenomenology is irrelevant;
that is, a phenomenological description of calculation leads us nowhere if, by
72 Thinking in Images

phenomenological description, one understands a description of symbols

and formulas used in the act of calculation. For the nature of the calculation,
it is irrelevant, for example, whether we use the symbol ‘1000’ or ‘103’. It is
also irrelevant whether the calculation is done with digits or words. What
distinguishes calculation from the description is not the use of digits, on the
one hand, and words, on the other. The term 2 + 2 = 4 can be written down
as ‘two plus two equals four’. What matters is the operation that is performed,
which means that to isolate the object to be explained, we have to separate the
relevant logical relations understood here in terms of a family of functions that
are carried out on a certain set.
If the ‘family of functions’ is an explanandum of the nature of the calculation,
what is the explanans? Notice that if we want to explain the nature of the ‘family
of functions’, we do not relate it to another, presumably more complex and
detailed family of functions. For example, trying to explain what function plus
is, we do not say that it consists of some more detailed parts, such as function
plus1 and plus2. In contrast, if we interpret the formula ‘a chemical element
with the atomic number 79’ as an explanandum, the explanans could describe a
more fundamental structure. It could be put, for example, in terms of a structure
consisting of quarks and leptons that are parts building larger structures. In other
words, a more profound explanation of something such as gold is a description
of a more detailed structure.
Understanding the nature of an operation, such as calculation, is quite
different. In this case, we need to understand the conditions that have to be met
for the operation to be carried out. For example, if we want to understand what
the operation plus is, we may point out instantiations of the function plus and
some set-theoretic assumptions that have to be accepted if the function is to be
logically possible. In this case, for example, we may adopt Zermelo-Fraenkel’s
axiomatic system. A more profound explanation of the operation would relate to
a richer conceptual scheme, for instance, to some second-order logic or category
theory axioms. Similarly, if we want to explain the interpretability of language,
we may adopt compositional semantics. Compositionality would function here
as part of the explanans of the interpretation function in an argument of the
following form: if language is to be interpretable, which is the case, then it has to
possess a compositional structure.
The same conclusion applies to the case of thinking, in general, and thinking
with images, in particular. First, to explain the nature of thinking with images,
one would have to designate an explanandum, isolating the functional roles of
 What Answers Should We Expect? 73

images in thinking. In other words, the first objective is to establish what images
do in the operation of thinking.
Second, after isolating an explanandum, the required explanation needs
to describe the conditions that must be met if images can perform certain
functional roles.7 Therefore, the first step is to investigate the functions of
images in cognitive processes to determine which functions are in accord with
the imagistic theory of thought. The point is to distinguish such operations
that cannot be carried out without images and distinguish them from non-
imagistic ones, such as operations requiring numerals and words to be
expressed. The second step is to establish conditions that have to be met if
images are to cover the requirements of the latter. It has to be established
how the concept of an image needs to be interpreted if images can play the
designated role. In short, to answer the question ‘What is thinking with
images?’, we must first ask what thinking with images does and then we must
try to find something that does that.8
The Traditional View holds that the answer to both questions is that images
represent by resembling objects. I will show that such an answer is not completely
wrong but not completely right either. In the following chapters, I show that
images are essential for the operations of constructing and recognizing objects.
Following that, I show that an image can represent the ways of constructing and
recognizing objects by exemplifying these ways.

Summary

To answer the question ‘What is thinking with images’, one has to determine
what kind of answers one expects. There are two strategies available. Neo-
Lockean theories hold that imagistic thoughts are theoretical objects introduced
to explain the relation between thoughts and perception. Imagistic thoughts are
abstract entities mediating between perceptual and discursive representations.
However, the problem of the relation between thoughts and perception
cannot be solved by invoking the concept of imagistic representation. There is
a thoughts–perception border, and no intermediary object can bring together
representations that possess mutually exclusive characteristics.
In contrast to Locke, Wittgenstein–Ryle’s sceptical argument holds that
thoughts cannot be kinds of objects. In particular, there is no vehicle we think
with; thus, the question is senseless. If there are no objects we think with, the
question of what characterizes thinking with images is deprived of content.
74 Thinking in Images

I argue that Wittgenstein–Ryle’s strategy is an attempt to explain away the

problem, not to solve it. I introduce an alternative strategy, according to which
we can understand the subject of the question in terms of operations based on
imagistic representations. It means that the question ‘What is thinking with
images?’ is equivalent to the question ‘What kind of operations are expressed
by images?’ Accordingly, I present a way to formulate a sensical and non-trivial
answer to it. It is put in terms of the operational approach, which means that
thinking is a matter of carrying out certain operations and explaining the
operations is equal to pointing out the conditions that have to be met if such
operations are to be possible. In a nutshell, I argue that to answer the question
‘What is thinking with images?’, one has to find out what images do and then
find something that does that.
 4

What do images do?

According to the operational approach, to understand what thinking with

images is, one has to understand what kind of operations images can perform
and find something that does it. The traditional answer to this question is that
images represent objects by means of resembling them. However, left as it is,
this answer is insufficient, for it does not address the issue of the irreducible
role of images in thinking. Moreover, it makes imagistic thoughts vulnerable to
epistemological, semantical and metaphysical objections described in Chapter 2.
In the following three chapters, I explore the irreducible roles of images,
looking at two case studies taken from mathematics and physics: a diagram of
a knot and a picture of a black hole. First, I argue that the way they represent
the world cannot be explained by resemblance-based and description-based
semantic theories; second, they demonstrate genuine semantical functions of
images. I hold that images represent the ways of constructing and recognizing
objects. I explain how to understand the operations of construction and
recognition, and, drawing on that, I introduce the two-dimensional model of
iconic reference and the measurement-theoretic account of images. In a nutshell,
I hold that images exemplify the rules of construction that enable us to identify
the depicted objects. The best explanation is that images should be taken as
measurement devices comparable to rulers and balances.

Case Study 1: Knot diagrams

Drawing on the case of knot diagrams, I demonstrate that one of the roles of
images is to represent the ways objects and events can be constructed. Let me
start by describing the knot theory.
Knot theory is a branch of topology dealing with knots and knot diagrams. A
knot is a smooth closed curve in a three-dimensional Euclidean space. For every
76 Thinking in Images

knot, there exists a knot diagram representing it (Cromwell, 2004, Theorem

3.3.2). A knot diagram is a representation of a knot with the height information
at the intersection points, which tells us which strand passes over the other. An
unknot is a knot representable by a knot diagram without crossings. A trivial
example of an unknot is a ring.
Knots can be transformed in various ways and still remain the same knot, so long
as they are not cut. Properties that hold throughout such deformations are called
invariants. Two knots are equivalent when one knot can be smoothly transformed
into the other without cutting, gluing together or allowing a self-intersection
during the transformation. The equivalence relation captures the intuition that
some topological properties of objects remain unchanged under deformations,
regardless of the changes in the geometric shape of the object. The main task of the
knot theory is to determine the equivalence of knots, as represented in Figure 4.1.
One way to determine the equivalence of knots is to demonstrate that two
knot diagrams can be transformed into each other by a sequence of three kinds
of moves, known as Reidemeister moves. According to Reidemeister’s Theorem,
two knots are equivalent if and only if there is a sequence of Reidemeister moves
transforming a knot diagram of one knot into the knot diagram of the other. This
means that for any two projections of the same knot, a sequence of Reidemeister
moves will transform one projection into another.
There are three types of Reidemeister moves (Figure 4.2). Type I allows us to
twist the knot or remove one while the rest of the knot remains unchanged. Type
II allows us to move one loop over another. Finally, type three allows us to move
a string over or under a crossing.
Applying Reidemeister moves is sufficient to determine the equivalence of
knots. For instance, we can demonstrate that the knots represented in Figure
4.1 are equivalent by applying the Reidemeister move of the first type to both of

Figure 4.1 The unknot (on the left) and an equivalent knot (on the right). © Piotr
Kozak.
 What Do Images Do? 77

Figure 4.2 Reidemeister moves. © Piotr Kozak.

the crossings. However, there is no known algorithm for applying Reidemeister

moves. It is just trial and error.
Computational methods in knot theory are based on the concept of knot
invariants. A knot invariant is a function that assigns a certain algebraic object
to each diagram of the knot so that the same values correspond to equivalent
diagrams. If we can find an invariant that takes different values for given diagrams,
then the diagrams are not equivalent. There are computational notations of knots,
such as the Dowker notation or the Conway notation, that help to solve the
problem of knot equivalence by applying the algebraic description of the knot.
Does this prove that algebraic ones can replace image-based methods?
Definitely not. Suppose that we can provide sufficient equations to solve the
problem of knots equivalence. Yet, what we want to know is not only whether
one knot can be transformed into another but also how it is done. We want
to know how to construct a knot, that is, what are the steps that lead from an
unknot to a knot. Metaphorically speaking, even if we know that a tie can be
tied, we still want to know how to do it.
Let me explain. Some formalizations can describe a diagram of a knot. For
instance, the Dowker notation can capture the first and second Reidemeister
moves. However, without the diagram, we cannot interpret these notations.
Consider the following example. Dowker notation represents the first
Reidemeister move by representing the number and character of the crossings
with consecutive integers. However, we still need to know how to perform
the first Reidemeister move. Without such knowledge, we would be left with
a meaningless mathematical formalization. We have to know what kind of
construction these numbers represent. Therefore, image-based operations are
irreducible in knot theory even if we can provide formal descriptions of the knots.
To illustrate it, let us consider the trefoil knot. The difference between the left-
trefoil and right-trefoil knots is well captured by two knot diagrams represented
78 Thinking in Images

Figure 4.3 The left-trefoil knot and the right-trefoil knot. Source: Wikimedia
Commons.

in Figure 4.3. However, this difference is not representable by Conway

polynomials. The Conway polynomial of each kind of the trefoil will be the same.
To capture this difference, it is necessary to introduce Jones polynomials. In this
case, diagrams are necessary to determine what constructions are representable
by a mathematical description.1
Diagrams are necessary to represent knot constructions for they exemplify
them (more on exemplification in Chapter 6). The theorem that every knot
is representable by a diagram should be read as follows: I can understand the
description of a knot construction only if I have the disposition to understand
a diagram exemplifying this construction. In contrast, I can understand the
diagram without knowing the corresponding mathematical description. For
instance, I can understand the description of the first Reidemeister move only if
I can identify the first Reidemeister move in a diagram. In contrast, I can identify
the first Reidemeister move in the diagram, without knowing its description.
Thus, understanding the description of constructions is built upon the ability to
understand constructions in diagrams.
To sum up, a knot diagram manifests how knots can be constructed. It shows
possible transformations of knots according to specific inferential procedures. This
aspect of knot theory is not reducible to the problem of knot equivalence and can
be represented by diagrams with well-defined operations, such as Reidemeister
moves. In the following sections, I explicate the concept of construction and
explain its implications for the theory of the content of diagrams.

What are constructions?

Let us reconsider the knot diagram represented in Figure 4.1. What distinguishes
this diagram from a scribble is that it has content, that is, it can be correct and
 What Do Images Do? 79

incorrect. However, there is no one predefined interpretation of this diagram.

There are infinite ways to get from the knot to the unknot. It can be twisted,
unfolded or its orientation changed, but it cannot be cut. What determines the
correctness standards of these interpretations?
To understand what a knot diagram represents, one has to know what procedural
steps are allowed by the diagram. De Toffoli and Giardino (2014) identify it with
the dynamic nature of diagrams. A dynamic diagram is a representation closely
related to specific inferential procedures that help define possible transformations
of the diagram (De Toffoli and Giardino, 2014; Hegarty, 1992). Our interpretation
of a knot diagram is based on the ability to imagine performing some operations
on the diagram according to some inferential procedures.
However, not all diagrams have a dynamic nature. A diagram of a triangle is not
dynamic. Moreover, we do not have to visualize transformations. We have to know
what kind of transformations are correct and incorrect. Imagining transformations
of the diagram can be a means to identify correct and incorrect transformations.
As I have argued, knot diagrams exemplify the ways objects can be
constructed. I call these ways ‘the rules of construction’ of an object. Let us
unpack this definition.
The concept of a rule is understood here as a function that applies correctness
standards to some construction procedures. Construction is correct if it is in
accord with a rule. Otherwise, it is incorrect. A rule describes some moves
as correct or incorrect, depending on whether they are allowed by the rule.
Importantly, knowing a rule means neither knowing its description nor the
propositional knowledge-that, such as knowing the content of an algorithm. It
is a kind of procedural knowledge that determines whether a move is correct or
incorrect.
The knot diagram in Figure 4.1 represents the rules of transforming a knot into
an unknot by applying the Reidemeister move of the first type. The Reidemeister
moves are transformations that are allowed by the construction rule, cutting any of
the knot’s crossings is not permissible. The diagram’s content is interpreted based
on recognizing these rules. In short, I know that the knot diagram represents the
unknot since I understand how to transform one into another.

Rules of Construction: functions ascribing correctness standards to construction

procedures.

How to understand the idea of exemplifying rules of construction? Consider the

case of a tangram (Figure 4.4), a game of putting together flat figures to arrive at
the desired pattern (Ertz, 2008).
80 Thinking in Images

Figure 4.4 A square pattern matched with a tangram set. © Piotr Kozak.

Note that tangram rules are exemplified by an outcome of applying these

rules represented in Figure 4.4. If someone asks how to play tangram, I can show
him Figure 4.4. In the same way, if someone asks how to build a house, I can
show him possible element arrangements in an architectural project. The way
we construct an image is inseparable from the outcome of this construction
displayed in the image. The image exemplifies the rules of construction.
Exemplification of construction rules is generalizable (Lehrer, 2000).
Suppose that I am asked to describe tangram rules. I can point to a particular
image exhibiting a pattern. Yet, this is not to say that I am looking for only this
particular image or the images that correspond to its description. It is essential
to understand how to go on. To understand the rules of construction, one has to
possess non-propositional knowledge of how to apply these rules.
Let us compare it to the way a ruler works. A ruler exemplifies a measure
by means of which you identify length. However, it does not mean that a one-
metre-long ruler can locate only one metre. Knowing how to use a ruler implies
that you can measure all distances.
There are two types of construction rules: the rules of interpretation and the
rules of production (see Hyman, 2006). Rules of interpretation determine how
to identify the represented object (more on the recognition-based identification
in Chapter 5). Rules of production determine how to create the representation
or the represented object. The ability to apply these rules is dissociated. I can
understand what is being represented in a diagram without having the ability to
create the diagram or the represented object. In the same way, I can understand
a foreign language without being able to speak it.
It is easy to confuse these rules. However, it is even more important not to
separate them. They are not different kinds of rules. They are different applications
of the same rules of construction. They both determine the correct and incorrect
 What Do Images Do? 81

moves to localize the represented object. In the same way, interpreting a sentence
and producing the sentence are governed by the same rules of grammar. Even
though these rules are applied differently, they can be easily brought together.
In the case of applying the rules of interpretation, it is essential to identify
in a representation the information to localize the searched object. In the case
of the rules of production, it is essential to identify the matching elements to
create this object. Let us consider a ruler. It shows you how to find the searched
distance and how to create a line segment. By the same token, a triangle diagram
can show you how to form a triangle (according to the rules of production) and
identify it (according to the rules of interpretation). In both cases, the diagram
represents the rules of construction of a triangle.
What is construction? It is usually interpreted as a procedure (or an outcome
of such procedures) of employing instruments, such as a ruler and a compass,
to draw a geometrical figure. This is called classical construction. However, in
the most general sense, employed by Euclid and Proclus, the concept refers to
a procedure of adding something that is ‘lacking in the given for finding what
is sought’ (Proclus, 1992, 159).2 Let me elucidate it with the help of a Euclidean
example.
Take three line segments a, b and c, which satisfy the condition that any two
line segments together are longer than the third. This is ‘the given’ in Proclus’s
description. Next, we construct a line segment DE, and we place on it the line
segment DF that corresponds to the length of a. Then we construct the line
segment FG that corresponds to b. We do the same with the line segment GH
that corresponds to c. Then we construct two circles. The first circle has a centre
F and radius FD. The second circle has centre G and radius GH. The two circles
intersect at K. It is now enough to join K, F and G to arrive at the construction
of ∆FGK (Figure 4.5).
Knowing the length of a, b and c and the definition of a triangle is insufficient
to solve the task of constructing a triangle. We need to know the rules of
construction for triangles. Even if we seek a polygon with three edges and three
vertices, we need to know the correct moves leading from the line segments a,
b and c to a triangle. Knowing the construction rules allows you to arrive at a
triangle if you are given three line segments and know what a triangle is. The
diagram represents these construction rules.
The concept of construction is reducible to neither the concept of the
construction elements nor the definition of the constructed object. It is possible
to get different triangles from the same definition and the same line segments.
In the case of Figure 4.5, you will get triangles with different properties, for
82 Thinking in Images

Figure 4.5 The construction of the triangle ∆FGK. © Piotr Kozak.

example, different heights, depending on which line segment is used as the base.
Construction is a skilful application of these elements in order to get to the
defined object.
According to Tichy (1986),3 the concept of construction can be generalized
and applied to non-geometrical objects. Following Tichy, I characterize
constructions as denoting the procedures of arriving at a goal (an object, event
or scenario). How does this apply to non-geometrical objects?
Let me use an algebraic example.4 Note that the term ‘7 + 5’ denotes the same
mathematical object as the term ‘20–8’, namely the number 12. However, these
terms denote 12 in different ways. Apart from naming 12, they identify different
procedures for arriving at 12.5
The construction ‘20–8’ consists of ‘20’, ‘8’ and the subtraction function. In
the same way, the construction of a triangle consists of the line segments and
the procedures of arranging them. The construction ‘20–8’ consists of the same
elements as ‘8–20’, including the same subtraction function. And yet, these
are different constructions, for the construction rules set a different order of
arguments. In the same way, by applying different construction rules to the same
elements, we construct different geometrical objects.
How to understand the concept of construction as a procedure for arriving at
a goal? Let us consider the case of projective geometry. It consists of a set of rules
localizing objects in space. However, projective geometry does not describe some
prefixed properties of some absolute space. It constructs a space. Geometric
projection does not represent localizations of some points, as if there were some
absolute space with fixed localizations. It represents the ways of localizing these
points by determining the parameters of the space. In Riemannian terms, to
determine the spatial properties of some manifold, we have to apply a geometric
measure. One of such measures is projective geometry.
 What Do Images Do? 83

Generalizing this example, constructions can be described as ways of localizing

the target by determining the parameters of some space. These parameters,
called here construction parameters, are (a) the basic units of construction, (b)
the order between (a), (c) and the relation between (a), where the basic units
of construction are interpreted in the most general sense as the states of some
informational space. In terms of the results of the operation, construction can
be seen as the ordered triple of the construction parameters‹ 〈a,b,c〉 that localize
the constructed object.

Construction (in terms of operation): χ is a construction of target T iff (i) χ

determines the parameters of some informational space, where these parameters
are (a) the basic units of χ, (b) the order between (a), and (c) the relation between
(a); in order to (ii) arrive at T.
Construction (in terms of the results of operation): an ordered triple of
parameters (a), (b), and (c), consecutively.

We can localize the states of two different spaces. The logical space is the set
of all possible states. Determining the parameters of the logical space is about
searching for possible distinctions among ways for the world to be as well as the
relations between these possible states. It is about searching for such pairs of
distinctions which coincide and such which are in contradiction.
The physical space is a set of states of affairs. Determining the parameters of
the physical space means searching for what the world is. Moreover, it is a search
for the relations between these states of affairs.
The logical space is closely related to the physical space, for determining the
parameters of the physical space is to single out one side of the distinction made
in the logical space (Rayo, 2013). For instance, to construct the proposition
snow is white, we isolate two possible states of the world – ‘white-snow’ and
‘non-white snow’ – and hold that the actual world overlaps with one side of this
distinction.
The simplest case of construction is localizing a point in the physical space.
Let us take a plane and try to locate a point on it. To do that, we have to determine
the values of the physical space and the relation between these values that will
establish the coordinate parameters. To find the localization of a point is to
determine these parameters. A Cartesian diagram exemplifies this construction.
It shows you how to find this localization. Interpretation of the diagram is based
on the reading of the construction parameters.
Notably, a point represents spatial locations and is determined by spatial
coordinates, but it is not something that is in space being projected onto the
84 Thinking in Images

plane. Space is made of points, but it does not imply that there is some space with
points in it. Space is a set of points. Something can be in some location in space,
but localizations are not in space in the same way as chairs are. Localization does
not have the property of being localized, as if there were some other localization
containing it. Localizations are in relation to other localizations, in the same way
as being a 5 is being in relation to other numbers, nothing less or more. To set
a point in some space is to set relations of this point to others. In this sense, to
determine the localizations of the points is to determine the properties of space.
The Cartesian diagram does not depict some prefixed space. It determines the
properties of the space to localize the point. Compare it to the way rulers work.
They do not represent some prefixed distances; they localize these distances by
determining the properties of the space.
Interpreting construction (in terms of the result of operation) is based on
localizing the construction parameters, that is, relevant sets of information, their
order and relation to other sets, to identify the searched object. Additionally,
constructions consist in abstracting away from irrelevant information. For
instance, the construction of a triangle identifies spatial properties, such as an
orientation and the length of line segments, and sets the geometrical relation
between these line segments. It determines the parameters of space to localize
some spatial structure, that is, a triangle. At the same time, it abstracts away
from irrelevant information, such as the colour and thickness of line segments.
Interpreting the triangle diagram is based on identifying the ordered triple of the
construction parameters.
By the same token, constructing 12 consists in identifying the values of the
logical space, arranging them in order, and applying a relevant function, such
as the subtraction function, to localize the value 12. Moreover, it abstracts away
from irrelevant information, such as whether apples or fingers are subtracted.
By constructing an object, we set the parameters of some space that determine
what is being represented. However, constructing objects has to be distinguished
from evaluating the truth values of representation. The construction ‘7 + 5’
localizes ‘12’, which, in turn, can be the basis of a proposition that 12 is a sum of
7 and 5. This proposition has true values. However, the very construction is not
truth-evaluable.
Compare this proposition to the operation of dividing by 0. It is not a false
operation. It is an operation that fails to determine the parameters of content.
Here, we do not ask whether the proposition identified by the operation of
dividing by 0 is true or false. We ask what value, if any, can be identified by such
an operation. Depending on whether the goal has been achieved, a construction
 What Do Images Do? 85

can be successful or abortive but not true or false. We can be wrong trying to
divide by 0, but we are not expressing a false proposition. We are applying an
operation incorrectly.
Constructions may seem to be truth-evaluable, for they can be put into
descriptive contexts. For instance, I can hold that it is true that you can construct
a triangle from three straight lines. However, knowing construction procedures
has to be sharply distinguished from knowing a description of a procedure.
Constructing an object is a matter of exercising a skill rather than knowing a
description of the object construction. For instance, applying Reidemeister
moves is a skill that requires practice rather than descriptive knowledge of what
Reidemeister moves are.
To stress this difference, let us compare constructions with algorithms.
Informally speaking, an algorithm is a description of a procedure for finding
a solution in a finite number of steps. Some constructions are algorithmizable,
for algorithms can represent constructions. However, constructions cannot be
reduced to algorithms.
An algorithm can be characterized as a finite set of well-defined instructions
on how to move from an input to an output effectively. Moreover, according to
the Church-Turing thesis, every algorithm is computable and implementable
in a Turing Machine. The most common way to implement an algorithm is to
express it with a recursive function. For instance, subtraction is based on the
algorithm made of some values, representing the algorithm’s input and the
recursive subtraction function. The subtraction function is applied to calculate
the output of the algorithm.
How does the concept of construction fit into this picture? Constructions
cannot be functions. Functions are elements of constructions. A function is a
procedure of associating the elements of a domain with the single elements of
the range of the function. However, a function can operate only if the elements
of the function’s universe are predefined. In other words, we must determine the
sets it operates on to apply a function. A construction localizes the function in a
certain informational space that we consider in a given situation.
Consider the operation of dividing by 0. The condition not to divide by 0
does not follow from the content of the divide function. Dividing by 0 is not a
wrong function, either. It is the same function as dividing by any other number,
but the function is applied incorrectly here. Dividing is a function that calculates
the number of times one number is contained within another. To know that we
cannot divide by 0, we have to notice that the application of the divide function
for the divider 0 leads to contradictory results. The condition not to divide by
86 Thinking in Images

0 was introduced to our understanding of the divide function since, without it,
we would get nonsensical mathematical constructions, such as 1 = 2. In other
words, the application of the divide function for the divider 0 is an abortive
construction, but it is not a wrong function.
Similar observations apply to algorithms. To use an algorithm, one must
construct it. For instance, it has to be decided what kind of a function has to
be used and what the sets representing the input and output are. Although the
construction of an algorithm is algorithmizable, it cannot be an algorithm since
then we would need an algorithm for the algorithm constructing algorithms and
so on.
In contrast, let us think of constructions in terms of procedural knowledge.
Learning constructions is a matter of acquiring the skill of applying a procedure
to localize a space state rather than knowing a description of this procedure. It
is a matter of exercising procedural knowledge based on recognizing parameters
of some space and not exercising propositional knowledge based on the
description of this space. That is why constructions have to be exemplified rather
than described.

The content of diagrams

With the concept of construction, we are now in a position to offer an approximate

definition of the content of diagrams. Further on, in Chapter 6, I will use this
definition to characterize the iconic content.
In general, diagrams show how to localize some objects in the logical
and physical space by determining the parameters of some informational
space. In particular, diagrams show what happens when certain construction
procedures are applied. In short, the content of diagrams can be characterized
by the construction rules of the represented object. For instance, a diagram
of a triangle shows that if you take three line segments and connect them
appropriately, you get a triangle. A diagram is correct if it aptly represents such
procedures. It is incorrect if the represented procedures do not allow you to
arrive at the goal.
Consequently, to correctly interpret a diagram is to identify the construction
parameters and construction rules that determine the permissible transformations
of the represented object. Grasping the content of a diagram is recognizing the
parameters of construction with respect to the rules of construction to localize
the searched object. For instance, to understand the knot diagram in Figure 4.1
 What Do Images Do? 87

is to identify the construction rule, which specifies that you can arrive at the
unknot if you apply the first type of Reidemeister move to the knot’s crossings
without cutting any of them. To understand a triangle diagram is to identify the
construction rule that specifies the construction parameters of the triangle. In
sum, the content of diagrams consists of the construction rules used to localize
a constructed object.

An approximate definition of the content of diagrams: diagram D represents target

T, iff D’s content C identifies (i) the construction that allows one to arrive at T;
with respect to (ii) the rules of construction.

The content of a diagram identifies the construction rules of the object rather
than denoting the object it stands for.6 Thus, two diagrams have the same content
iff they express the same construction rules instead of denoting the same objects.
The terms ‘7 + 5’ and ‘20–8’ represent different constructions, although both
denote 12. The number 12 is individuated by different constructions.
The most obvious illustrations of this definition of content are architectural
drawings, manuals and maps. The standards of correctness for the architectural
drawing of a building are determined by such construction rules that allow us to
erect it (according to the rules of production) and identify it (according to the
rules of interpretation). An incorrect architectural drawing is such that it does
not allow us to construct or identify a building. Interpretation of the drawing
is based on identifying the construction parameters and the construction rules
that determine the permissible ways of transforming these parameters. Manuals
teach us how to build something step-by-step and how to identify the relations
between the elements of the mechanism. A correct manual is the one that takes
us from an initial state to the goal state and helps us to identify the represented
mechanism. A map is correct if it aptly represents the ways we can localize some
points in some space. A correct interpretation of the map is based on identifying
these ways.
To get a broader picture, consider two less-intuitive examples. The line graph
displayed in Figure 4.6 represents inflation growth. It identifies the inflation rate,
mapping it on time. These are the construction parameters.7 Inflation growth is
an abstract object localized by construction procedures setting the partial order
between the parameters of the inflation rate and time. The graph represents
inflation growth by identifying the target, which is an increasing function
associating the inflation rate with time. Understanding this graph is identifying
the construction parameters and the rule that if you map the inflation rate on
time, the inflation rate will eventually rise. The diagram localizes the inflation
88 Thinking in Images

Figure 4.6 Line graph representing inflation growth. © Piotr Kozak.

Figure 4.7 The pulley system diagram. © Piotr Kozak.

growth in relation to the inflation rate and time. Vertical and horizontal axes
indicate relevant information. The irrelevant information is the colour of the
line.
Next, consider the pulley system diagram represented in Figure 4.7. It
shows how the mechanism behaves if the string is pulled down. The correct
interpretation of the diagram is based on identifying the construction
parameters: the information on the orientation and direction of the string as
well as the relation between them. The visual shape of the weights is irrelevant
information. Understanding this diagram consists in grasping the rule that if the
string is pulled down, the weight goes up. The movement of the string can be
visualized. Yet, the visualization is only a means to grasp the diagram’s content.
Understanding the pulley system diagram is about knowing how the system
behaves if you pull the string.
Thus, the inflation and pulley system diagrams represent constructions.
They show that if you identify the basic construction units and connect them
 What Do Images Do? 89

according to the rules, you arrive at the target. The correct interpretation of these
diagrams is based on identifying these rules.
One caveat is in order. There is a difference between Figures 4.6 and 4.7. The
inflation diagram represents specific data and the relation between them. It
represents a construction rule for particular values. It is equivalent to a diagram
of a particular token, such as a diagram of a particular triangle. Let us call it a
‘token-diagram’.
In contrast, the pulley system diagram represents a general rule of how a
system behaves if the string is pulled down. It abstracts away from particular
values of weight or shape. It represents a general construction rule for building a
type of mechanism. It is equivalent to a diagram of a triangle as such. Let us call
it a ‘type-diagram’.
Every diagram can be interpreted as either a type-diagram or token-
diagram.8 For instance, an architectural drawing can show how to build
a particular building, but it can be taken as showing how to build a type of
building. The inflation diagram can be taken as a type-diagram if it represents
the general rule of how inflation rises over time. The pulley system diagram
can be taken as a token-diagram if interpreted as representing a particular
mechanism.
Knowing how to construct a type-diagram does not imply constructing a
token-diagram. I can know a general rule for constructing triangles without
constructing any particular one in reality. Moreover, a token-diagram can never
express the generality of type-diagram content.
However, every type-diagram requires the existence of a token-diagram to
be represented. For example, to show how to construct triangles as such, I have
to use a triangle token-diagram. Having a type-diagram implies that I have the
disposition to construct and recognize the token-diagram of a relevant kind.9

Properties of constructions

A construction procedure has to be distinguished from the constructed object,

for there can be constructions that construct no object. Nothing is constructed
when dividing 12 by 0. Yet, it does not mean there is no construction of
dividing 12 by 0. It is an abortive construction that has no denotation. In this
sense, it is a different construction than ‘12–12’, which denotes 0. An abortive
construction fails to single out its referent. The construction ‘12–12’ referent is
an empty set.
90 Thinking in Images

This implies that constructions’ properties differ from the properties of

constructed objects. The ‘12–12’ construction consists of 12 and the subtraction
function. The number 0 denotes an empty set. Yet, 0 does not consist of 12
and the subtraction function. If the content of the number 0 consisted of the
subtraction function and all possible values of the function, then it would have
to consist of infinitely many kinds of functions that lead to 0. Instead, applying
the subtraction function is the way of arriving at 0.
Consequently, knowing the content of a diagram is knowing the construction
rules, not the properties of the constructed object. Knowledge of the construction
rules for an object is irreducible to the knowledge of any description of that
object. It covers the Euclidean idea that knowing how to construct a triangle
is irreducible to knowing the definition of a triangle. Moreover, this explains
why one can learn how to identify triangles without knowing what triangles are.
Once they know how triangles are constructed, children can distinguish them
from squares without knowing that triangles are polygons with three edges and
three vertices.
The misidentification of the properties of constructions with the properties of
constructed objects gives rise to the false belief that our thoughts are composed
of the word- or image-like objects. Words and images are the means by which we
construct thoughts. That is the idea expressed by Wittgenstein–Ryle’s sceptical
argument.
At the same time, this does not mean that there exist some intramental
objects, such as thoughts, that can be described by words and images. Thoughts
are abstract objects, such as numbers, which are the effects of construction and
do not exist outside the construction context. The question ‘where are thoughts’
is as nonsensical as the question ‘where are numbers’.
Moreover, the misidentification of these properties makes us vulnerable to
falling into the pictorial fallacy. Images show how objects can be constructed
and how they would look if they were constructed in some way. Yet, this does
not imply that they describe these objects.
Investigating the form of constructions is cognitively productive.
Manipulating the form of constructions can lead to the discovery of their novel
properties. Seeing that a + b = b + a cannot be inferred from the content of the
symbols. Discovering the commutative property of addition entails the capacity
to recognize that changing the order of the elements does not affect the result of
the construction.
At the same time, constructions are characterized as operations determining
the parameters of some informational space. Thus, different constructions can
 What Do Images Do? 91

highlight different aspects of information, enriching our understanding of

constructed objects.
Let us consider a geometrical example. Note that the Euclidean condition that
any two line segments of the triangle together are longer than the third is not
analytically derivable from the definition of a triangle. The content of the definition
– ‘a polygon with three edges and three vertices’ – does not contain the postulate
that any two line segments together are longer than the third. Euclid introduces
this postulate, for, without it, the construction of the triangle is impossible. In other
words, we know that this postulate is part of the content of the triangle concept
because, without this postulate, it would be impossible to construct a triangle.
Thus, investigating the way in which mathematical objects can be constructed
teaches us something about their nature. That is why diagrams can be useful
in the context of discovery in mathematical practice. Particularly, interpreting
the content of diagrams in terms of the rules of construction helps us to
clarify the concept of aspect shifting in diagrams. Note that diagrams can be
rearranged, which makes them open to new interpretations. The same diagram
can be arranged differently, which allows for addressing its various properties.
Giaquinto (2007) recognizes this as a crucial aspect of visual thinking in
mathematics. For Peirce, this is an iconic property of diagrammatic reasoning
(CP 2.279). For instance, manipulating the construction of triangles helps to find
a proof strategy for Pythagoras’ Theorem.10
However, it is not clear what these different aspects of mathematical objects
are. According to the view presented here, they can be identified by means of
different constructions. Different interpretations of a diagram single out different
construction parameters.
Let us consider the well-known Necker cube, an ambiguous figure with
two different interpretations. We can understand the content of these two
interpretations in terms of displaying different constructions. We can pick out
the lower-right and upper-left faces and arrange them according to the ‘being-
in-front’ relation. We get two different constructions depending on the order
of the arguments in this relation. In the same way, we can have two different
constructions for the subtraction function and arguments ‘20’ and ‘8’.
Thus, different aspects of information can be identified by using different
constructions. For instance, line and bar graphs (Figure 4.8) use the same data. Yet,
they are differently constructed, affecting the interpretation of the information.
The line graph (a) makes it easier to identify the relationship between exercise
and weight loss. An increasing function connects the data. In contrast, bar
graph (b) makes it easier to single out the relationship between exercise and the
92 Thinking in Images

Figure 4.8 Line graphs (a) and bar charts (b) convey the same information but in a
different way, affecting the accessibility of the information. © Piotr Kozak.

number of calories burned because the bars comparing the data for calories and
the amount of exercise are closer to one another. The line and bar graphs employ
the same data but construct it differently. The graphs highlight different aspects
of information, depending on the applied construction.

Explaining resemblance

In his Syllabus (CP 2.277), Peirce defines diagrams as the representations

that refer to relations via resembling them. Although this characteristic is, in
principle, not false, it is a source of a deep misunderstanding.11
It is far from obvious what we mean when we say that diagrams resemble
anything. The resemblance is usually understood as sharing first-order properties
 What Do Images Do? 93

of objects. The diagram representing inflation growth does not resemble inflation
growth. Inflation growth is an abstract object that cannot be a relatum in the
relation of resemblance.
At first glance, we can try to explicate the concept of resemblance in terms
of structural similarity. The concept of structural similarity denotes sharing
the same structure in the following sense. Two sets, A and B, are structurally
similar iff there is a function mapping (some) elements of A and (some) relations
defined over the members of A onto (some) elements of B and (some) relations
defined over the members of B such that preserves the second-order properties
of the relations in A. Thus, if some relation holds for (some) elements of A,
the corresponding relation holds for (some) elements of B (e.g. Kulvicki, 2014;
Swoyer, 1991).12 Consequently, two objects, a and b, are structurally similar if
and only if various relationships among the parts of a correspond to important
relationships among the parts of b.
Structural similarity is not a sufficient condition to determine the content of a
diagram. A trace left by an ant can be structurally similar to Winston Churchill,
but it does not refer to Winston Churchill (Putnam, 1981). Yet, structural
similarity seems to be a necessary condition of being a diagram and having
content.
To illustrate the idea of structural similarity, let us consider Euler diagrams.
The core idea is that we can represent subset, intersection and disjoint relations
between sets in terms of inclusion, overlapping and disjointness of circles
(Figure 4.9). It may seem that the concept of resemblance is crucial to explain
why we can represent abstract relations by Euler diagrams. According to this
view, Euler diagrams represent logical relations because they mirror the logical
structure of these relations.
However, the concept of structural similarity is not necessary to explain how
Euler diagrams work. We can explain it by pointing out that they represent
the same construction rules. For instance, we can represent the subset relation
in terms of the inclusion of circles as they represent the same constructions.

Figure 4.9 Euler diagrams representing the subset, intersection and disjoint relations,
respectively. © Piotr Kozak.
94 Thinking in Images

The subset relation is constructed according to the same rules as the inclusion
of circles in the following sense. You can arrive at subset A if you include all
elements of subset A into set B. In the same way, you can arrive at the inclusion
of circles if all elements of one circle are contained within another.
Explaining how Euler diagrams work in terms of sharing the same
construction rules has a significant advantage over explaining diagrams in terms
of resemblance. It allows us to avoid the following categorical fallacy. Note that if
one uses the resemblance relation to explain the fact that the subset relation can
be represented in Euler diagrams by the inclusion of circles, one holds that the
subset relation and the inclusion of circles are identical in some respect.
However, that is nonsense. Subsets do not share any non-trivial properties
with the circles represented in Euler diagrams. Subsets and drawings belong to
different ontological categories. Subsets are logical, while drawings are physical
objects. Claiming that they are alike is as much nonsense as claiming that
numbers share some properties with apples. Apples can be red, numbers can be
even, but not the other way round. Numbers can represent the quantity of apples,
but it does not imply that they are alike. By drawing circles, we can represent
subsets, but it does not imply that the representation and the represented object
share some common properties.
Seemingly, the categorical fallacy objection does not apply to the structural
similarity theory. After all, it does not hold that logical objects, like relations
and numbers, and physical objects, like drawings and apples, share some first-
order properties, like being red or tasty. The idea is that these objects share some
second-order properties. Thus, a subset is structurally similar to a drawing of two
circles as some second-order properties, such as being an asymmetric relation,
are preserved. Second-order properties do not characterize the objects of first-
order properties. An apple can be red (first-order property), red is a colour
(second-order property), but an apple is not a colour. Therefore, the ontological
fallacy argument is misplaced.
Still, this gives us an unattractive ontology of mathematical objects. Compare
two sets, A and B, where B has the first-order property F of being a subset of A,
and two objects, a and b, where b has the first-order property G of being inside
a. The structural similarity theory holds that a and b can represent these two
sets in the following way. First, b has the property G, and G has the property H
of being a relation such that all elements of b are included in a. The object b is
not a relation, hence, it does not have the second-order property H. Second, set
B has the property F, and F has the property H of being a relation such that all
elements of B are included in A.
 What Do Images Do? 95

This story suggests that there is some logical object, ‘a subset’, that can be
described by the relation of being included. This does not seem right. The term
‘being included’ is only another way of saying that something is a subset. The
inclusion relation is not the property of being a subset as if there were some
logical objects described by this property. The inclusion relation is a way of
isolating some elements of a certain kind in a set. The effect of such a procedure
is a subset. But a subset is nothing more than ‘being included’. Something cannot
be a subset without being included.
In contrast, b is an object that can be identified by the property of being
inside a. However, b is not equal to the property G of being inside a. We use the
property G to identify b. Yet, the object and its property are distinct. The object
b could have no property G.
What is the moral of that? Note that the story told by the structural similarity
theory is plausible only if we adopt some sort of mathematical Platonism
regarding the nature of mathematical objects. It does not have to be wrong. It is
not necessarily correct either. However, it would be unfair to insist on adopting
Platonism only for the purposes of explaining how Euler diagrams work.
These ontological problems are avoided if we explicate Euler diagrams
in terms of representing the same construction rules, for the properties of
constructions are not the properties of the constructed objects. Euler diagrams
and logical relations can share the properties of constructions without sharing
the properties of objects. We do not have to be Platonists, which is fair.
Moreover, resemblance-based theories give us the wrong idea of the
epistemology of diagrams. First, it is not clear how resemblance-based accounts
can distinguish between misrepresentation and non-representation (Suárez,
2003). To misrepresent something is to assign the represented object with
properties that it does not have. A non-representation does not assign any
property. According to resemblance-based accounts, representing something is
sharing first- or second-order properties. However, if something does not share
any property with the represented object, it is not a representation. This seems to
be clearly wrong since misrepresenting a circle with the diagram of a square does
not imply that a diagram of a square is not a representation.
We cannot simply bypass this problem by invoking the idea of preserving
a partial structure (e.g. Bueno and French, 2011), for this idea is too inclusive.
For one thing, there is a partial structure-preserving relation between a
representation of a dot and a diagram of a circle and a square. Yet, it does not mean
that the single-dot diagram represents circles and squares. For another, there is
no way to distinguish between partial representation and non-representation.
96 Thinking in Images

A scribble does not represent anything, yet it preserves some partial structure
with any object.
Second, the concepts of resemblance and structural similarity are too weak
to identify the represented objects. In order to know what is being resembled,
we have to identify relevant aspects of resemblance. Yet, according to the
well-known Wollheim’s argument (1998, 2003), to identify relevant aspects of
resemblance, one has to identify what is being resembled. In Wollheimian terms,
recognizing an object depicted on a picture’s surface presupposes that we have
a seeing-in experience of this object. Provided that we have such experience,
we can recognize the depicted object in the marks on the picture’s surface. If
seeing-in were not prior to the experienced resemblance, there would be no way
to discern accidental and non-accidental marks on the surface.
Consider an ellipse-shaped diagram. It is structurally similar to a circle seen
from a perspective. Without knowing what this diagram represents, we cannot
know which parts of the structure are relevant. However, it was by comparing
structures that we were supposed to identify the object of reference.
At first glance, this should not be particularly disturbing. Most of the
proponents of resemblance-based theories acknowledge that sharing
properties is not a sufficient condition of representation. It is held (e.g. Abell,
2009; Blumson, 2014; Bueno and French, 2011; van Fraassen, 2008) that the
epistemological problems can be resolved by supplementing the condition of
sharing properties with the pragmatical condition that the representation has
to be used by an agent in a proper way. The general idea is that to recognize
the relevant aspect of resemblance, one has to recognize the intention of the
agent.
However, such a move only replaces one problem with another. Note that we
shift the burden of the argument from resemblance-based accounts to intention-
based accounts, which have their own problems, and, as Wittgenstein’s argument
from content indeterminacy shows, cannot explain depiction.
Moreover, such a shift renders the resemblance relation functionally
irrelevant. The resemblance does nothing in the process of recognizing the sign’s
content, for all we have to do is to recognize the agent’s intentions. Without any
supplementation, the intention-condition is insufficient.
Explaining the content of diagrams in terms of the rules of construction
not only gives us a better picture of the content. It avoids the epistemological
problems of resemblance-based accounts and provides us with an explanation
of what resemblance is. Most of all, it allows us to make sense of the concept of
resemblance without sharing first-order or structural properties.
 What Do Images Do? 97

Let me illustrate this with the following example. Take two diagrams
represented in Figure 4.10. The dashed line represents unfinished lines. Note
that they lack similarity in shape and structure since some lines are missing.
However, they represent the same construction rules, thus having the same
content. For it is possible to construct a triangle if the straight lines are drawn
further in thought. The fact that we run out of space on paper plays no role in
constructing triangles.
In contrast, two identical ellipse-shaped diagrams can denote two different
objects: either an ellipse or a circle seen from a perspective. These diagrams have
the same spatial properties but different content, for different construction rules
are applied.
Next, depending on the kind of identified constructions, we can determine
the relevant aspects of resemblance. Consider two triangles ∆ABD and ∆ACE
in Figure 4.11. Do they resemble each other? It depends on the rules of
construction we recognize as relevant. Suppose we want to know whether they
are geometrically similar. In that case, we are interested in the construction rule,
which identifies as relevant the information on the ratio of the sides’ lengths
and the measures of the angles. If we want to know whether they are congruent,

Figure 4.10 Two diagrams of a triangle with the same construction rules. © Piotr
Kozak.

Figure 4.11 Depending on the identified rules of construction, ∆ABD and ∆ACE are
similar or congruent to each other. © Piotr Kozak.
98 Thinking in Images

the construction rules single out the information on shape and size. ∆ABD and
∆ACE are similar or congruent depending on the identified construction rules.
Consequently, the diagrams ∆ABD and ∆ACE represent the same or different
objects. In the most general sense, any two token-diagrams of triangles are
similar if they depict the construction rules for any polygon with three edges
and three vertices.
Finally, the idea of construction gives us an insight into Gombrich’s so-called
riddle of style. According to Gombrich (1960), adopting the resemblance-based
view of depiction does not allow us to address the question of why people at different
times and in different cultures tend to depict objects in different ways. Gombrich
argues that this riddle is based on a false assumption that depiction is based on
capturing a visual likeness of the depicted objects. Based on this assumption, pictorial
representational systems should have developed in one direction, perfecting the
ability to express the resemblance so that eventually, there should be one style that
unifies all others and represents the world most adequately.
To address the riddle of style, Gombrich acknowledges that depiction is based
on recognizing familiar objects in patterns of lines and colours, which elicits an
illusion of seeing these objects. To recognize objects in these patterns, we have
to employ schemata, that is, stereotypic ways of depicting objects. Schemata are
a means to represent the world in the same way as projection in cartography
serves to represent certain regions in the world. Different schemata are used to
represent different aspects of the world, just like different methods of projection
are used to represent different aspects of space.
It is rarely noticed that the same riddle of style applies to diagrams.13 If
diagrams were based on structural similarity, then we might expect a tendency
to create a unified diagram form that would preserve structures most adequately.
That contradicts the common observation that there is a reverse trend in creating
many different forms of diagrams.
The concept of construction covers Gombrich’s intuitions without applying
the vague illusion-metaphor. Let us compare the line graph and the bar chart
displayed in Figure 4.8. They both represent the same structure of information.
Yet, depending on their construction, the aspects of information they represent
are different. We expect that different styles of diagrammatic representations
will coincide with the different constructions they represent, which seems to
agree with our observations. Moreover, the illusion-metaphor can be explicated
in terms of sharing the properties of constructions but not the properties of
objects. That seems to align with Wittgenstein’s and Frege-Davidson’s argument
of lack of truth values.
 What Do Images Do? 99

Where do we stand with the concept of resemblance? Much depends on how

we understand resemblance. It is an umbrella term with shifting reference being
insensitive to counterexamples. It can be understood in relational and non-relational
terms (e.g. Hyman, 2012), as a symmetrical and asymmetrical relation and so forth.
However, saying that resemblance cannot be an explanation of how images
work is not saying that we can dismiss resemblance as an insufficient condition of
how images represent; or that images do not share properties with the represented
object. I hold that images do share a structure with the represented objects.
Moreover, I hold that resemblance plays an essential role in the process of
iconic representation, and the fact that an image resembles the represented object
has to be explained by any full-fledged theory of depiction. If our explanation
of depiction were unable to explain the concept of resemblance, it would be
considered insufficient; for the concept of resemblance is the intuitive starting
point of any theory of depiction (Schier, 1986; Walton, 1973). It is enough to say
that the inability to explain how images can resemble their objects is the chief
reason why the Goodmanian theory of depiction seems to fail.
The theory of construction can explain how diagrams resemble represented
objects. The resemblance between any two objects can be explained in terms of
representing the same construction rules. Although the concept of resemblance
cannot explain how diagrams work, the concept of construction gives us an idea
of how diagrams can resemble their objects and have content. In a nutshell, the
resemblance relation does not determine the construction rules. On the contrary,
the construction rules determine the resemblance relation. Diagrams resemble the
relations in terms of representing the same construction rules. To determine what
is being resembled, one has to determine what the rules of construction are.

Summary

Based on the analyses of the content of diagrams, I demonstrate that diagrams are
necessary for representing the construction operation. I explicate the concept of
construction in terms of procedures of arriving at a goal by means of determining
the parameters of some informational space. Accordingly, I distinguish between
the properties of constructions and the properties of objects.
I describe the relationship between the concept of construction and
resemblance. I claim that the presented description of content is able to explain
the concept of resemblance without taking this concept as an explanans of the
content of diagrams.
100
 5

Recognition-based identification

In the previous chapter, based on the analysis of diagrams, I pointed out that
images exemplify the ways of constructing objects. In this chapter, I discuss the
second role of images: triggering the ability to recognize objects and events.
Drawing on Evans, I introduce the concept of recognition-based identification.
I argue that images can convey a reference relation distinct from demonstratives
and descriptions. It can be explicated based on the construction concept. I
hold that recognition is based on identifying the construction invariants which
determine the properties of some space in order to localize the depicted object.

Case Study 2: The picture of a black hole

When the Event Horizon Telescope (EHT) caught its first glimpse of the M87
black hole, located over fifty-four million light years away from Earth, one could
almost hear a moan of disappointment. After all, did we not have more detailed
pictures before? Simulations of black holes had looked more stunning. Is the
blurry picture of a hazy cosmic flame loop all that we can get?
The picture represented in Figure 5.1 may not seem like much, but it is a
breakthrough in physics. Why does having this picture matter?
The black hole picture helps us validate theories and measurement methods.
Up-to-date simulations of black holes based on Einstein’s theory of general
relativity have been commonly used in science to predict how a black hole
behaves. As Einstein’s theory predicted, the shape of a black hole would be
almost a perfect circle, and its size should be directly related to its mass. As it
turned out, it is. The picture has allowed us, for the first time, to test Einstein’s
theory of relativity for black holes and other regions containing dense matter.
Furthermore, the picture is an object of investigation in that we now have a
way to study black holes as physical objects rather than mathematical constructs.
102 Thinking in Images

Figure 5.1 The first picture of a black hole. Credit: Event Horizon Telescope
collaboration et al. National Science Foundation.Source: European Southern
Observatory (ESO) via Wikimedia Commons.

Thanks to the picture, we are able to calculate the black hole’s mass more precisely.
Using the EHT, we are able to measure the radius of M87’s event horizon directly
and calculate the hole’s mass. Consequently, the calculation helps us validate
available methods of mass estimation.
Additionally, the picture shows us the features that equations have not
yet captured. For example, there is the open question of why the black hole’s
silhouette is irregular in shape. It is a question that we hope to answer in the
future. It is also believed that the data gathered by the EHT will offer some
insight into the formation and behaviour of black holes. The structure revealed
in the image can help to reveal how material gains the energy needed to fall into
a black hole or spiral away in jets. Lastly, it is believed that information provided
by the EHT will help us resolve fundamental physics problems, such as the black
hole information paradox.
Thus, the black hole picture is much more than a blurry illustration of a night
sky. Let us think of it as the final result of a measurement operation. The picture
represents measured features of the object, where the measurement values can
be projected onto the measured object. We study the features of the picture to
learn something about the features of the black hole. Before the picture was
taken, some facts were unknown and could not be deduced. If they could be
deduced, physics would be reducible to metaphysics (e.g. Elgin, 1996a, 2017).
However, the fact that the picture may seem underwhelming to some teaches
us something important about our understanding of what images are. Let me
illustrate it with Dennett’s (1990, 1991a) argument from computer reasoning.
 Recognition-based Identification 103

The black hole had been first measured and described in the numerical system
of representation, which, in turn, was presented as the black hole picture.
Displaying the information on the picture does not change the content of the
information. Before we displayed the information on the properties of the black
hole, we had had the information.
Moreover, based on the syntactic properties of numerical data, we can compute
the non-measured properties, such as the mass of the black hole. Displaying the
information via the picture does not affect the content. Therefore, having the
black hole picture is inessential for knowledge and reasoning. To show what
is wrong with this argument, let us turn to the idea of the naturalness of icons.

Naturalness of icons

The backbone of Dennett’s argument is the belief that depictions and numerical
representations carry the same kind of content since both code the same
information. The only difference is the way the information is displayed.
This belief is based on Goodman’s idea (1976) that images and language-like
representations are different kinds of a conventional representational system.
However, this does not seem right, for it fails to explain the properties of
the content of images. In contrast to language-like representations, images seem
to bear some natural relation to the represented objects, which explains why
a depiction of a horse looks like a horse. The language-like representation of
a horse does not have to be like a horse.1 What does it mean to bear a natural
relation?
The notion of ‘natural relation’ is deeply ambiguous. In the sense that
interests us here, it refers to a nonconventional relation. There are, however,
two understandings of the convention concept. The first holds that a sign
is conventional iff it has arbitrary semantics. The second holds that a sign is
conventional if it belongs to some representational system. It is noncontroversial
that images are conventional in the second sense, for they can be easily classified
according to different representational styles (Willats, 1997). It is highly
controversial whether they are conventional in the first sense. The concept of a
natural relation will be understood here as bearing a non-arbitrary relation to
the represented object.
The nonconventional relation between a sign and the represented object
can be explicated in two ways. First, it can be put in psychological terms (e.g.
Burge, 2018; Giardino and Greenberg, 2015). According to the psychological
104 Thinking in Images

understanding of this term, a representational system is more or less natural to

the degree to which natural psychological abilities, rather than cultural factors,
make the system easy to use. For instance, it is easier to represent directions
with left and right arrows instead of letters A and B, for the arrows better fit our
psychological abilities.
However, the psychological understanding is not of the appropriate modal
strength to explicate the idea of a non-arbitrary relation. It is metaphysically
contingent matter what kind of psychological abilities we possess. If we were
robots, it could be more practical to internalize letters instead of arrows. Such
defined naturalness is only the complement of the metaphysical arbitrariness of
the representational systems.
Second, naturalness can be understood in metaphysical terms. The
representational system is natural only if the properties of the system depend
on the properties of the represented object. The psychological naturalness of
icons can be a simple consequence of the metaphysical naturalness so defined.
What does it mean for the representational system to depend on the represented
object? Let me explain it with the help of the concept of natural generativity.
According to the recognition-based theories of depiction (e.g. Lopes, 1996,
2005; Sartwell, 1991; Schier, 1986; Squires, 1969), the identification of depicted
objects is generated naturally. The concept of natural generativity refers to the
observation that if someone has interpreted any member of an iconic system of
representation, then one can interpret any other member of the system, provided
that one can recognize it in perception. As Wollheim puts it (1987, 77): ‘if I can
recognize a picture of a cat, and I know what a dog looks like, then I can be
expected to recognize a picture of a dog.’ Knowing the meaning of one part of an
iconic system implies knowing how to interpret the meaning of the other parts.
The idea of natural generativity marks the difference between iconic and
symbolic systems. Symbolic systems lack natural generativity. They are based on
arbitrary semantical and syntactic rules that have to be taught one by one. Thus,
the knowledge of a symbolic system cannot be derived from the knowledge of
a particular part of the system. Knowing that German ‘Hund’ means a dog and
how cats look does not imply knowing what ‘Katze’ means. I do not understand
what ‘Katze’ means if I did not encounter this word before. In contrast, I can
understand pictures I have never seen before.
According to Schier (1986), any theory of depiction should be able to explain
the natural generativity of pictures. The recognition-based theories can provide
such an explanation granted that recognitional capacities triggered in picture-
perception are the same as in the case of perception in the flesh. Pictures cause
 Recognition-based Identification 105

naturally generated interpretations by activating the recognitional skills of the

competent spectators. In short, picture P depicts an object O only if P elicits
competent spectators’ capacities to recognize O. We can understand pictures for
the ability to understand them piggybacks on the everyday recognitional skills,
exploiting the resources normally deployed to identify objects in perception
(Lopes, 1996, 2003; Schier, 1986).
What can we learn from recognitional theories of depiction? Drawing on
Schier, I hold that the irreducible role of images is to trigger the recognitional
capacities of the interpreter. A successful interpretation of a picture involves the
capacity to recognize the depicted object. We see objects through their depictions
(e.g. Aasen, 2016; Lopes, 1996, 2005; Walton, 1984; Wollheim, 1987). Symbols
cannot do that. Images convey a special kind of reference relation which is not
accessible for symbolical representations. Thus, images are necessary for the
operation of recognition. The fundamental question of any theory of depiction
is how to explain this reference function (Budd, 2008; Newall, 2011).
Let us go back to the black hole picture. Unlike numerical data, the picture
makes it possible to refer numbers to physical properties and recognize them in
the physical space. The picture offers us a direct link to the investigated object,
mapping the logical space of numerical data onto the physical space. The black
hole picture triggers recognition abilities and enables us to relate abstract models
to physical reality.
However, that cannot be the whole story. First, suppose we follow Schier
and interpret recognition in psychological terms as a set of perceptual abilities.
In that case, the recognition-based theory is useless when explaining how we
understand the black hole picture. We cannot see a black hole in the flesh. The
black hole picture cannot trigger the same perceptual skills as in the case of
perception.
Second, various theories of depiction are compatible with the claim that
pictures activate the same recognitional capacities as in the case of perceiving
the depicted object in the flesh. The question is how pictures can do so.
The most intuitive alternative to recognition-based theories are resemblance-
based theories. According to resemblance-based theories, the most natural
explanation of the fact that pictures trigger recognitional capacities is that pictures
resemble the depicted objects, where the resemblance is usually understood as
sharing first- or second-order properties with the represented object (Hopkins, 1998;
Hyman, 2006; Neander, 1987; Newall, 2011; Novitz, 1988). However, resemblance-
based theories cannot explain how we can recognize a black hole. Determining
the resemblance relation presupposes the ability to make a comparison between
106 Thinking in Images

the relata. In the case of the black hole picture, such a comparison is impossible.
Pictures are commonly used to depict unobservable phenomena (Schöttler, 2017).
Resemblance-based theories cannot explain how this is possible. Thus, we need a
better (and non-psychological) explanation of recognition.

What is recognition?

The concept of recognition-based identification was introduced by Evans (1982)

to distinguish this kind of reference from demonstratives and descriptions.
Recognition is an operation that makes it possible to reidentify an object or
event. To explicate the recognition concept, let me start with some distinctions.
Let us consider seeing the black hole in the picture. We can distinguish
between three types of perception here. First, we can perceive the black hole.
Second, we can recognize the black hole. Third, we can see that it is a black hole.2
Drawing on Hope (2009), I call these object perception, perceptual recognition
and that-perception.
All three types of perception are functionally distinct (Overgaard, 2022). I
can see a black hole, but I do not have to see that it is a black hole. I can mistake
it for a picture of a donut. Seeing a black hole does not entail recognizing a black
hole. I can see something without recognizing it. Recognizing a black hole does
not entail seeing that it is a black hole. I can recognize the picture of a black hole
as similar to other pictures without seeing that it is a picture of a black hole.
Object perception is basic to others. I cannot recognize a black hole and see
that it is a black hole without perceiving the black hole. Object perception can be
described as the ability to fix perception on some objects and scenarios.
Object perception consists in extracting a perceptual object from the
background and distinguishing it from other perceptual objects, a process
that underlies perceptual selection and tracking of objects (e.g. Casati, 2015).
In order to do that, the representational system has to identify the perceived
object demonstratively. For instance, it has to be determined that the texture
of a picture’s surface is irrelevant to object perception. Next, the system has to
determine topological relations to set part–whole relations between perceptual
objects. For one thing, this consists in distinguishing the whole object from the
background and the other objects; for another, in identifying the parts of the
whole object.
Misrepresentation of object perception consists in the inability to identify the
object. For instance, I may not be able to distinguish between the features of the
 Recognition-based Identification 107

depicted object and the features of the picture’s surface or between the object
and the background.
Perceptual recognition describes the ability to associate perceptual objects
with each other and reidentify an object after some changes. It is often described
as recollecting information and having a sense of familiarity with the recognized
object (e.g. Dokic, 2010; Hope, 2009; Hummel, 2013). Here it is understood as a
perceiving that perceived objects belong to the same kind.3
Misrecognition can be easily confused with errors in that-perception. For
instance, looking at the black hole picture, I can misinterpret it as a picture of
a donut. Yet, errors in perceptual recognition are distinct from errors in that-
perception. The inability to recognize two objects as belonging to the same kind
is different from the inability to identify the kind they belong to.
Perceptual recognition is a non-propositional, non-conceptual and non-
linguistic form of perception. A child can recognize triangles without knowing
that they are polygons with three edges and three vertices or that they are called
‘triangles’. In order to do that, the child must learn that some spatial properties
remain the same across changes in a different context.
This ability is non-propositional. Children can be wrong in confusing
triangles with squares, but they are not expressing any false proposition. It does
not require concepts (e.g. Siewert, 1998). It is, rather, a prerequisite for having
concepts (Peacocke, 1992; Giovannelli, 2001). Before one can subsume an object
under a concept, one has to be able to recognize this object. Recognition is a
gateway from object perception to cognitive processes such as categorization
and reasoning, but it is not based on these processes.
Only that-perception requires the presence of language-like, conceptual and
propositional structures (e.g. Armstrong, 1997). If I see that it is a black picture,
it must be possible to express it in the language. It is conceptual since it involves
subsuming a representation under a concept. It is propositional since perceiving
states of affairs is truth-evaluable. If I am wrong when I see that it is a black hole,
then it is not true that it is a black hole.
Object perception, perceptual recognition and that-perception of the black hole
use the same kind of perception. Seeing-that is based on sensory perception, but
it is not another kind of perception. It is not ‘perceiving the truth of propositions’
as if there were a perception of propositions. Propositions are ways of identifying
how the world can be. These are not the objects we can perceive. Seeing-that is,
rather, a form of propositional attitude based on perception. Accordingly, I cannot
see that it is a black hole without having a belief that it is a black hole. Seeing that
it is a black hole and believing that it is not a black hole is self-contradictory.
108 Thinking in Images

In contrast, I can recognize a black hole without having a belief that it is a

black hole. I can recognize my mother in a medieval painting without believing
that it is my mother. However, in accord with Russell’s Principle, I cannot have
that-perception without the recognition of the object of perception. I cannot
see that it is snowing without the ability to recognize snow (e.g. Evans, 1982;
Peacocke, 1992).

Varieties of reference

According to the influential account in cognitive science, perceptual recognition

is connected with indexing an object with an object-file (e.g. Green and Quilty-
Dunn, 2021; Murez and Recanati, 2016; Pylyshyn, 2007; Recanati, 2012). An
object-file is a representation that sustains reference to an external object
over time. It helps to explain how minds can keep track of objects in different
contexts. This ability is illustrated by the well-known ‘tunnel effect’ (Burke, 1952),
according to which if an object moves beyond a visual obstacle and reappears on
the other side, it is perceived as a single entity. This is true even if the reappearing
object has different physical properties from the disappearing one. Thus, there
are strong reasons to believe that the information stored in the object-files is
dissociated from the information on the object’s features. Subjects can reliably
track objects despite significant changes in colour, shape or size, which implies
that object-files have directly referential (nondescriptive) semantics.
The object-file framework seems to offer an elegant explanation of how
recognition works. The story goes as follows (e.g. Recanati, 2012, 85 ff.):
during the first perceptual encounter with a target, we form a demonstrative
file that indicates the target. Next, the demonstrative file is converted into a
memory demonstrative which stores information on past encounters with the
target. Recognition is based on finding an index that converts the memory
demonstrative into the recognitional demonstrative, which identifies the link
between the perceptual and memory information. Recognition consists in being
reacquainted with a target.
This story is simple but uninformative without explaining how memory search
works. First, the cognitive system cannot automatically identify the memory
file, for it would have to search through all the possible files, which would soon
lead to the combinatory explosion. The search cannot be based on looking for
similarities between two demonstratives, as in the case of Gauker’s (2011) idea of
perceptual similarity spaces, for it would require the object-file to be descriptive.
 Recognition-based Identification 109

Such files would have to contain information on the features of the target, which,
in turn, would make them unable to explain why the ability to track objects is
disassociated from the information on the object’s features. Moreover, it does
not solve the combinatory explosion problem without any explanation of how
looking for similarities works. Interpreting the idea of similarity as primitive and
unanalysable does not offer an explanation but excuses.
Second, the proponents of the object-file framework can interpret recognition
in terms of identifying probabilistically determined causal links in the following
way. If we systematically encounter target T producing demonstrative file D(t),
there is an increasing probability that it would result in producing associated
memory file M(t). Recognition is a matter of causally driven activation of D(t)
and associated M(t).
However, that won’t do. For one thing, the causal story is vulnerable to
counterexamples. I can recognize objects I did not encounter in the past,
for example, I can recognize my mother in her childhood photograph. I can
recognize objects I have a wrong idea of, too. If it turned out that previous
visualizations of a black hole were wrong, and the depicted black hole picture
resembled a square rather than a circle, scientists would still identify it as a black
hole picture.
Moreover, the story is empirically implausible, for recognition abilities seem
to be functionally dissociated from the ability to retrieve information from
memory causally. I can remember perceptual objects without recognizing
them. For instance, patients with associative visual agnosia can successfully
copy drawings but cannot recognize the copied objects (e.g. McCarthy and
Warrington, 1986). Next, I can recognize a familiar scenario connected to a false
memory of myself as a child being lost in a supermarket. False memories are not
causally connected with any events.4
Thus, although the object-files framework can work well to explain the
mechanisms of demonstrative reference, it does not seem to explain the nature
of recognition-based identification. This suggests that recognition is not a kind
of demonstrative reference.
It seems natural to think that recognition involves perception. Consequently,
it may seem that recognition-based identification is a kind of demonstrative
reference (e.g. Strawson, 1963). Demonstratives are based on a causal
information link between a representation and its target and the ability to locate
and track this target in space. This implies that to understand the content of
a demonstrative, one has to identify a relevant causal link and find the target
within some egocentric frame of reference, determining the here-now relation.
110 Thinking in Images

Indicating a target means determining the spatiotemporal relation between

myself and the indicated object (e.g. Evans, 1982; Kaplan, 1989; Matthen, 2005).
Recognition-based identification cannot be a kind of demonstrative
(Bermúdez, 2000; Curry, 1995; Zeimbekis, 2010). Recognition-based
identification does not put anything within the egocentric space. The target of
the black hole picture is not ‘there’ as if indicating the picture would localize the
black hole. The picture’s target is not located within the here-now framework,
implying that we can recognize it without having a demonstrative of the target.
For instance, we can recognize Santa Clause in a picture without being able to
refer to Santa Claus demonstratively.
The source of the confusion is that we can demonstratively refer to the
representation of the target. For example, I can indicate my mother’s photograph
and hold ‘that is my mother’. Yet, it is not demonstrative of a target. If I place a
picture of my mother next to her, I am not holding that my mother is in two
locations at once. I refer to my mother by recognizing her in the picture.
However, recognition-based identification is akin to demonstratives in at
least one respect. Recognition seems to differ from identification by description,
which holds that we identify objects by matching them to their descriptions.
However, the descriptive theory of recognition seems empirically and
conceptually implausible.
It is empirically unsupported, for it cannot explain how we can track objects
that change their features across temporal and spatial contexts. Notably, it
faces well-known objections against template and feature theories of pattern
recognition in cognitive psychology (e.g. Liu et al., 1995). The template theory
holds that we recognize perceived objects by matching them to a template
stored in long-term memory. However, it does not explain how we can store the
information about recognized objects. We perceive objects constantly changing,
which would require storing infinitely many templates of perceived objects.
The feature theory holds that we recognize objects by matching the selected
features of objects. However, it cannot explain how we can recognize different
interpretations of bistable objects which share the same features. Moreover,
object recognition skill, called ‘o’ skill in cognitive psychology, is dissociated
from general intelligence and propositional knowledge. Neither IQ nor SAT
scores can predict the recognition of objects (Richler et al., 2017).
The descriptive theory is conceptually incoherent, for I can recognize an
object without knowing its description. For instance, I can recognize my mother
in her childhood photographs without knowing what she looked like. Moreover,
I can know a description of the object without any ability to recognize it. For
 Recognition-based Identification 111

instance, I can know what my aunt looks like but cannot recognize her in a
photograph. The recognition-based identification seems to refer to the target
directly. Recognizing my aunt in a photograph differs from knowing a description
‘the woman represented in the photograph is my aunt’ (Lopes, 2010; Terrone,
2021a). The question is what such recognition-based identification involves.

Construction invariants

The core idea of recognition is identifying a target by fitting it into a pattern.

This pattern does not have to be recollected. It can be discovered. Recognizing a
pattern is finding an invariant. It is seeing that a target remains the same across
changes and contexts, including changes in viewpoint and target properties. For
instance, I can recognize my mother in her childhood picture if I identify some
transformation invariants, that is, the properties that remain unchanged after
transformations. Depending on the context, this invariant can be a birthmark, a
number of fixed points and so on.5
Failures in recognition consist in the inability to find such invariants. For
example, perceptual recognition can be impaired in patients with apperceptive
visual agnosia who cannot recognize known objects in different contexts. They
have difficulty recognizing common objects viewed from unconventional angles
but have no difficulty identifying objects shown in conventional orientations
(Warrington and James, 1988).
The crucial point is to explain what these invariants are. The notion of
invariance was introduced into the theory of depiction by Gibson (1971) to
explain how pictures can convey the same optical information as in perception.
The concept of optical information covers the idea that when one sees an object,
one does not see only its front surface but the whole of it: the back and the front.
In a sense, all of its aspects are present in the experience.
According to Gibson (1973), perception is a matter of identifying formless
and timeless invariants that specify the object’s distinctive features. We can
think about perceptual invariants in the same way as we think about geometrical
objects. The form of a geometrical figure, like a triangle, is the face of the figure.
While the face can change during transformations, the substance of a triangle
remains the same.
A picture is an array of persisting invariants of a structure. They can be
perceived not only when the perspective keeps changing, as in ordinary
perception, but also when it is arrested, as in pictures. This means that when I
112 Thinking in Images

see a cat, I perceive an invariant of the cat. When I encounter a picture of this cat,
I am prepared to pick out the relevant invariant. This is not to say that I see an
abstract cat, I have that-perception of a cat, or I perceive common features of the
class of cats. What I get is information on the persistence of structure.
A picture is a record of perception. It enables the observed invariants to be
stored. These invariants can be retrieved to convey knowledge. Thus, pictures are
an efficient method of teaching and learning. However, the knowledge pictures
convey is not explicit and language-like. According to Gibson (1978, 1979), the
formless invariants cannot be put into words. They can be captured but not
described.
However, the invariant concept is unclear (Fodor and Pylyshyn, 1981; Gombrich,
1971). Particularly, it is not clear what it means that some invariants can be perceived
when the perspective is arrested. Let us consider the projective geometry case. In
his early account of depiction, Gibson (1954) defends the projection-based theory,
according to which a picture conveys optical information through the geometrical
projecting of three-dimensional scenes onto two-dimensional surfaces. In short,
picture P depicts object O iff there is a systematic mapping between P and O
such that it preserves some properties determined by the projection invariants.
Based on the principles of projective geometry, interpretation of P is a matter of
identifying the scene that was projected onto the picture.
The projection-based theory seems to fit well with satellite maps, photographs
and Alberti’s paintings. However, Gibson (1971) explicitly rejects this theory.
First, it cannot explain distorted pictures like Cezanne’s paintings. Second, it is
at odds with depiction practice. Contrary to popular belief, artists have never
paid much attention to projective geometry clarity and abstract elegance. For
instance, they commonly confuse the habit of putting the vanishing point in
the centre of the picture, which is a matter of composition, with the perspective
projection system as such (Gombrich, 1972). Third, the ability to recognize
depicted objects in the flesh survives significant changes in the properties of
the representation and the represented object (Lopes, 1996). For instance, we
can reidentify depicted objects, even if the depiction does not represent the
perspective correctly and the depicted object has changed over time. This implies
that the idea of projective geometry is too restrictive to explain recognition
(Inkpin, 2016).
However, projective geometry gives us an insight into what invariants are. Let
us consider the cross-ratio invariant represented in Figure 5.2.
The cross-ratio remains unchanged across projective transformations. It is
preserved regardless of the shape of the projection plane. It is the invariant and
 Recognition-based Identification 113

Figure 5.2 The cross-ratio invariant. The relation between points is preserved in
projective transformations in such a way that the cross-ratios AC / BC: AD / BD and
A′C′ / B′C′:A′D′ / B′D′ are equal. © Piotr Kozak.

the property of the projection system. However, it would be wrong to say that it
does not describe the properties of the depicted object. It certainly does. Based
on the lengths of A′C′, B′C′, A′D′, B′D′ we can deduce the lengths of AC, BC, AD,
BD.
The main idea is that images are representations of invariants. We can observe
the invariants in the projection plane and apply them to the represented reality
(e.g. Elgin, 2010a). In this sense, we see invariants in images. How should we
understand this?
The concept of construction is useful here. The content of diagrams has been
provisionally defined in terms of representing construction rules. A successful
interpretation of a diagram is based on identifying construction parameters
determined by these rules. The same definition is extendable to all image genera.
Constructions have been characterized in terms of determining the parameters
of some space in order to arrive at the constructed object. By determining
these parameters, we determine the properties of some informational space.
The general idea is that some properties remain unchanged across different
constructions. These properties are the construction invariants. They determine
what is being preserved during different acts of determining parameters of a
space.

Construction invariant: the property that is identified as unchanged in the act of

construction.

Let me go back to the projective geometry example. If we apply the rules of

projective geometry to construct the line segments ABCD and A’B’C’D’, the
cross-ratios are preserved. The cross-ratio is the construction invariant. It means
114 Thinking in Images

that whatever the construction parameters, such as the vanishing point or the
lengths of the line segments, are, the cross-ratio remains the same. We can
identify the represented object length based on the lengths of the projected one.
The idea, however, is that projective geometry is just a special case of the more
general construction operation. As Riemann might say, projective geometry is
only one of the many possible constructions. In other words, projective geometry
is only one way of determining space parameters.
To explain the concept of construction invariants, let us consider the terms
‘7 + 5’ and ‘20–8’. They represent different constructions that identify different
parameters of the logical space. However, these constructions pick out the same
target. The number 12 is no less than the invariant of different constructions
(while not being a projective invariant). We identify 12 by finding that this value
remains unchanged in various constructions.
Next, think about triangles different in shape and size, represented by
two diagrams. The diagrams exemplify different constructions. However,
these constructions preserve some properties, such as the number of angles.
Interpretation of these diagrams is based on identifying these construction
invariants. Based on that, we can identify the constructed objects as belonging
to the same kind.
Depending on the identified construction invariants, we can recognize
different targets. For instance, if we construct a triangle whose invariant is the
sum of its angles, then we identify different shapes of triangles as representing
one kind. If the construction invariant is the angle measure, then we identify two
triangles with the same angles as similar.
Cezanne’s paintings present a more demanding case. To explain it, let me go
back to the Gibsonian concept of perceptual invariance. It describes a familiar
phenomenon of perceiving objects as stable throughout environmental changes.
For instance, perceived colours seem relatively constant under changing
illumination conditions. A red apple looks red at midday and at sunset. This
subjective constancy of colours helps us to identify objects and is indispensable
in object recognition.
According to Gibson (1979), the (approximate) invariance of the colour
experience cannot be a matter of identifying the perceiver-independent
properties of a perceptual object, such as the light wavelength. These properties
differ throughout the changes in illumination and perspective. Instead, colour
constancy is a matter of tracking relational features of the perceptual object. It
is a skill of identifying the perceived colour in relation to other colours across
contexts (Buccella, 2021; Green, 2019).
 Recognition-based Identification 115

Figure 5.3 Munker-White’s Illusion. Although the grey bars A and B depict the same
hue, B appears to be brighter. Source: Wikimedia Commons.

A good illustration of the relational nature of colour constancy is the so-called

Munker-White’s illusion (Figure 5.3). The grey bars, A and B, depict the same
colour. However, we perceive them differently, depending on the colour of
the bordering stripes. Our perception of colours is partly determined by the
surrounding colours.
According to the relational colour constancy hypothesis (Craven and Foster,
1992; Davies, 2016; Foster, 2003, 2011), recognizing the perceived colour by the
perceptual system cannot be based simply on picking the colour out. The system
has to contextually identify the relations between different colours to find the
searched one. It must track the colour of the target surface by identifying the
relations of this colour to the adjacent surfaces as the illumination changes. In
other words, the system has to identify the parameters of some space of colours,
such as the colour of the adjacent regions surface, and learn how they change across
different perceptual contexts. The searched colour is invariant across different
parameters of the perceptual space of colours, including changes in illuminations
and perspective. The colour constancy mechanism is based on keeping track of
invariant relations among adjacent surfaces as the perceptual conditions change.
What does it teach us about Cezanne’s paintings? One of the impressionists’
main goals is to depict how we perceive colours. To do so, Cezanne, using
paint, determines the parameters of the depicted colour space to find the
searched colour. He picks out some colours and identifies relations between
them, determining the construction parameters. He achieves the goal if such
constructed colour space preserves the same colour as in perception. The colour
experience is the invariant of this construction.6
I hold that recognition-based identification consists in identifying the
construction invariants connected with knowing the construction rules that
116 Thinking in Images

determine possible transformations of the constructed object. Recognition-

based identification means knowing what the permissible transformations of the
object are and what is preserved during these transformations. Let me explain it
starting from some examples.

Recognition-based identification: subject S recognizes target T, iff (i) S identifies

the construction’s invariants of T; and (ii) the construction rules of T.

Recognition-based identification can be a simple operation. Suppose you are

looking for a colour hue for your bedroom. To identify the searched colour
you can take a colour chip (note that it is not a projective geometry case). We
can change the chip’s size and shape. However, the colour property has to be
preserved. Based on the chip, you can recognize the searched hue. The chip is the
representation of the colour invariant by means of which you identify the colour.
Similarly, to identify the way we experience colours, you can look at Cezanne’s
paintings. Although they do not preserve shapes and distances, they tend to
capture how the colours are perceived. The colour experience is the construction
invariant. Cezanne’s painting is a representation by means of which you identify
how one perceives the colour.
Notably, recognition-based identification is directly referential. It is not a
matter of knowing an object’s description and matching properties. To know
the reference of the construction 7 + 5, you do not have to compare 12 to some
mysterious searched value, looking for similarities between them. You directly
identify 12 as the correct answer because you know how to count.
Perceptual recognition is a skill of identifying the parameters of some
perceptual space – basic representational units, like edges and colours, and
their relations – and transformational principles. Recognizing that perceptual
objects belong to the same kind is a skill of identifying invariants that order
some perceptual manifold. Invariants are the means to identify structures and
objects in perception (Ison and Quiroga, 2008; Lowe, 2004).
Images are representations of these invariants. Images do not show objects.
They show how to localize them by identifying construction invariants. They
show how to see objects.
Comparing it one more time with the structural theories of depiction can
make it more clear. According to structural theories (e.g. French, 2003; Kulvicki,
2014), images are structure-preserving representations that are structurally
similar to the represented objects. Let us recall the following definition of
structural similarity: two sets, A and B, are structurally similar iff there is a
function mapping (some) elements of A and (some) relations defined over the
 Recognition-based Identification 117

members of A onto (some) elements of B and (some) relations defined over the
members of B such that preserves the second-order properties of the relations
in A. For instance, sets A and B are structurally similar if they are isomorphic
without necessarily sharing any first-order property.
Now, structural theories of depiction are not wrong. Moreover, the theory
of construction invariants can be seen as a version of a structural theory. A
structural correspondence between the representation and represented
object is another way of saying that some object properties are preserved
by the representation. For instance, a map identifies a territory only if there
is a structural correspondence between the map and territory. However, the
structural theory heavily depends on how we understand the condition ‘some’.
If we want to find these elements and relations, we need a defined domain, range
and mapping function. It is no problem in the case of mathematical operations.
It is a problem, however, if we want to map a defined mathematical structure
onto an undefined set of states of affairs. Structures are not some Platonic
objects waiting to be discovered in reality and compared. It is not sufficient to
say that semantics is fixed or that some classes or mappings are more natural.
It is nothing more than a metaphysical postulate (Isaac, 2019; van Fraassen,
2008).
The image goal is to mark the elements and relations to impose order on some
manifold and identify the structure. Images determine the properties of the
informational space to localize the searched object. Successful interpretation of
an image is based on finding the construction parameters (together with knowing
construction rules) that help us to localize this object. For instance, the triangle
diagram highlights the construction parameters, like the number of angles, in
order to identify some objects as belonging to the same kind. Understanding the
diagram implies that you know that you can change the diagram’s size and shape
(but you cannot disconnect the elements of the structure), and some properties
are preserved. It does not involve comparing it to some ideal Platonic object. It
involves knowing how to identify triangles by learning what remains unchanged
across different constructions.
Let me illustrate it with the following example (Figure 5.4). Suppose you have
some manifold represented by a set of points (Figure 5.4a). Images impose order
on this manifold by determining the construction parameters: a relevant set of
points and order relations (Figure 5.4b). Interpretation of the image is based
on identifying these construction parameters. Knowing that this construction
preserves the number of angles, we can identify it as representing a triangle
(Figure 5.4c).
118 Thinking in Images

Figure 5.4 Images impose order on some manifold to localize the elements of the
structure. © Piotr Kozak.

Are construction invariants the properties of the representational system

or the represented object? It is a misplaced question. Recognition is the skill
of identifying the construction invariants to localize the searched object.
Construction invariants are the ways of describing the objects, not the
descriptions of an object. They show how to localize an object. Images exemplify
the construction invariants by means of which you identify some properties.
What does it mean?
Let us think about construction invariants as measurement invariants.
Measurement invariants are these properties of the measurement that indicates
that the same construct is being measured. For instance, the standard metre
length exemplifies the measurement invariant by means of which we measure
length. No matter what the world is like, the standard metre will be one metre
long. Does the standard metre describe the world? In the most general sense,
it is. There has to be a structural correspondence between the length of the
standard metre and the measured length. However, it should be understood as
saying that the length of the standard metre is taken to localize the properties
of the world. The standard metre represents the way of localizing the world’s
properties.
Now, you cannot ask how you know that the standard metre is one metre
long, for it sets the condition of identifying what one metre is. To understand
what the standard metre indicates, you have to learn what properties it preserves.
It is learning a measuring convention. Nothing more. By the same token, asking
how you know what construction invariants are preserved is the wrong kind
of question, for they set the conditions of identification. To understand what
construction invariants types are preserved is to ask about the convention they
are a part of. Nothing more. Think about a colourful triangle diagram. To ask
 Recognition-based Identification 119

how you know that it preserves shape properties and not the colours is like
asking how you know that a ruler represents a distance and not a ruler’s colour.
Thus, knowing how to localize objects is not based on feature comparison
or looking for the similarities between some objects. It involves knowing what
properties remain unchanged during different constructions. Consequently, we
can track and reidentify objects with different physical properties, for the object
description can change across different ways of localizing searched objects. The
invariant can be the mere property of being individuated by some construction.
Let me illustrate this with three cases.
First, consider a simple case of setting a localization of some point in space.
It can be done by determining the coordinates in relation to some frame of
reference represented by a Cartesian diagram. However, the frame of reference
and, respectively, the coordinates can be altered. Yet, when changing the frame
of reference, we do not change the localization of the point. We change the
description of the localization. This does not imply that the point is localized in
some absolute space. Here, localization is the invariant of the way we localize the
point. The Cartesian diagram determines the properties of this space.
Second, in the tunnel effect experiments, subjects are not asked to identify
the properties of perceived objects. They judge that the object that disappeared
when moving beyond the occluding shape and the one that reappeared on the
other side are one and the same even if all object properties have changed. Thus,
identifying likenesses is not necessary for recognition-based identification. In
the case of the tunnel effect, identifying the relation to the vanishing point and
the motion vector as a transformation invariant is sufficient to recognize these
objects as belonging to the same kind.
Third, the represented properties do not have to be ‘localizable’ in the image
(Wollheim, 2003). I can see that a depicted person is happy even though no single
property of the image can be identified as representing happiness. Happiness
is the depicted object’s property localizable by identifying construction
invariants. I can see-in the depicted person’s happiness by identifying patterns of
representation, for example, the composition of colours and lines in a painting.
These patterns are the construction invariants represented in a picture through
which we identify the represented object.
Let us go back to the black hole picture. Its interpretation involves determining
the construction parameters of the black hole. It is based on isolating the object
from the background by determining the parameters of some space. Recognition-
based identification of the black hole means identifying the pattern of the black
hole. It is not based on comparing the picture’s properties to the black hole, which
120 Thinking in Images

is physically impossible. It is about knowing which properties of the construction

of the black hole remain unchanged. For instance, determining the black hole’s
spatial properties means identifying the emission ring’s crescent shape and the
central shadow. These are the invariants of the black hole construction and the
means to localize the properties of the black hole. These properties are close
to the construction properties of a donut. That is why it is easy to confuse the
black hole picture with the picture of a donut. Yet, no one says that knowing the
interpretation conditions implies that we will always arrive at the correct one.

Iconic convention

Recognizing construction invariants is not an easy task. It requires expertise,

as in the case of knot diagrams and fMR images. The ability to recognize such
invariants is a matter of skill. Yet, when it is mastered, it enables us to interpret
different constructions that preserve the same invariants.
Consider the triangle diagram. If you learn how to identify invariants in
one triangle diagram, you learn how to identify them in any other. You learn
a convention of how to identify certain invariants. Learning this convention
implies that you know how to go on, in the same way as knowing how to count
implies that you know how to count to infinity.
Accordingly, iconic convention can be defined as a set of construction rules
that preserve the same type of construction invariants. Two pictures belong to
the same convention if the same invariants types are identified.

Iconic Convention: a set of construction rules distinguished by the same

construction invariants types.

For instance, projective geometry is a set of construction rules that preserves

cross-ratios. Distances are preserved in isometric transformations. Photographs
preserve optical information and light direction. Impressionist paintings
preserve the visual experience of the object and so on.
Knowing how to interpret images is knowing the iconic convention
determining the construction invariants type. For instance, reading a topographic
map is based on the ability to identify the information on distances and spatial
relations between locations. Distances and spatial relations are construction
invariants of the topographic convention. Alternatively, reading a topological
map is based on identifying topological relations, not distances. Topological
relations are invariants of topological constructions.
 Recognition-based Identification 121

The fact that there are many iconic conventions is not accidental. If
constructions are characterized in terms of determining the parameters of some
informational space, then different constructions can highlight different states of
the space. Iconic conventions can be compared to different systems of reference
that identify different invariants. If there is no absolute point of view, then there
can be no one iconic convention.
Interpreting iconic convention in terms of construction invariants types helps
us to address the problem of natural generativity. The problem is how to explain
the ability to interpret a whole system of representation based on the meaning
of its part. Applying the concept of the iconic convention makes the concept of
natural generativity explainable and avoids the problems of recognition-based
accounts of depiction. Consider a case of a mathematician who learns how to
interpret a graph of the function f(x) = x + 1. He learns the graph construction
rule. Knowing this, he can recognize other functions, such as f(x) = x + 2,
displayed on a graph, too. This seems to be the case of natural generativity.
However, there is no need to be perceptually acquainted with any function. It is
sufficient to know what properties it preserves.
Consequently, we can address the problem of the naturalness of icons.
According to the metaphysical understanding, the representational system
is natural only if the properties of the system depend on the properties of the
represented object. In this sense, iconic representations are natural. For instance,
if the distances between represented objects were different, then the distances
displayed on the projection plane would have to be different, too, for the cross-
ratio is preserved in projections. If Cezanne had a different visual experience,
then the visual properties of his paintings would have to change, too, for the
experience properties are preserved in impressionist paintings. In contrast,
symbolic representations are not natural. Whether the symbol A represents the
left or right direction is an arbitrary choice.
In contrast to recognition-based theories, knowing an iconic convention
does not involve perceptual acquaintance with the represented object. It is
sufficient to know which properties are preserved by the iconic convention. This
seems justified since we commonly depict objects we have never seen before.
For instance, we have no problems depicting fictional objects like dragons.
We do not have to see dragons in the flesh. Recognition-based identification
does not involve comparing two relata: the depiction and the depicted object.
It is sufficient to know how a dragon would look if it existed. If we know what
properties are preserved by the iconic convention, we can infer what kind of
properties dragons would have.
122 Thinking in Images

Let us go back to the black hole picture. Recognizing the black hole in the
picture does not require that we stand in some perceptual or resemblance-like
relation to the black hole. If we know the iconic convention of the picture, we
know which properties of the black hole it preserves. That is why we can infer
the properties of the black hole from the picture.
Consequently, the skill of recognition can be dissociated from having
descriptive knowledge. Let us suppose that the picture of a black hole resembles
a square rather than a circle. Still, knowing the iconic convention that optical
information is preserved in a photograph, we could recognize it as a black hole
picture. The square-like black hole would not correspond to our common way
of representing black holes, but no one says that science should align with how
we commonly represent things.

Summary

This chapter discusses the recognition concept and its relation to images.
Recognition is a natural and direct relation to the represented object. I hold that
it is a distinct kind of reference relation that is irreducible to demonstratives and
descriptions. Images are necessary to represent this reference relation.
I explain the recognition in terms of identifying the construction invariants,
that is, the properties that remain preserved in different constructions. I hold
that recognition is based on identifying these invariants. In the last section, I
explicate the concept of iconic convention, characterized as a set of construction
rules that preserve the same invariants.
 6

What is an image?

In the previous chapters, I characterized images’ functions as necessary for

representing operations of construction and recognition. In this chapter, I
present a two-dimensional model of reference for images. I hold that the best
way to understand the properties of this model is to think of images in terms
of measurement devices, such as rulers and balances. I hold that images show
how a measured object would look in some measurement set-up by representing
the ways of identifying this object. Finally, I apply the measurement-theoretic
account to explore the nature of impossible images.

Two-dimensional model of iconic reference

Before characterizing the model of iconic reference, some conceptual distinctions

are needed. Drawing on Cummins (1996) and Greenberg (2018, 2021),1 I
introduce a distinction between the referent, content and the representation
target.2 The target is the object or the state of the world the representation is aimed
at. The referent is the object or the state of the world that is actually identified by
the content. The content are the properties of representation through which we
identify the referent. The properties of the content determine the referent. The
target can be fixed even before the representation is created.
Distinguishing between the target and referent is essential for making sense
of the concept of accuracy conditions. The general idea is that a referent can
mismatch a target. For instance, I intend to draw a horse, but because of my
lack of skills, the drawing resembles a cow. A cow is a referent identified by the
content, and a horse is the target of this representation. If the referent mismatches
the target, an image is inaccurate. It is accurate otherwise.
Images can represent the same target but have different content. Suppose my
seven-year-old son’s drawing and Alberti’s paintings depict the same targets. Yet,
124 Thinking in Images

they depict them in different ways, having different content. There can be cases
when pictures have the same content but different targets, too. The same picture
of John can represent John, as well as a man as such if it figures in a biology
textbook.
Seeing Marilyn Monroe on black canvas (Schier, 1986) is a special case when
we identify a target without matching content. We take Marilyn Monroe to be a
target, but the content of the black canvas is not Marylin Monroe. The Putnam’s
ants case (1981) is the reverse of that. A Churchill-like trace left by an ant does
not refer to Winston Churchill. He is not a target of this representation. However,
the content of the trace identifies the properties of Winston Churchill. There can
be targetless representations that have content, too. Impossible images aim at
no target. Still, the impossible fork and the Penrose stairs have different content.
In depiction theories, these dimensions are often conflated. Wollheim’s (1987,
1998) seeing-in theory fails to distinguish between the content and target.
Consequently, it cannot identify correctness standards that determine whether
the viewer has correctly perceived what a picture represents. Wollheim suggests
that the author’s intentions can do the job (Wollheim, 1980, 1993). However, the
concept of intention can be applied to identify the target but not to the content.
My seven-year-old son and Alberti can have the same intentions, but they
produce different content (Hyman, 2012, 2015).
The resemblance-based theories are the theories of content, not target.
Resemblance can determine correctness conditions for content but is too liberal
to specify a target. The reason is that, in some respect, everything resembles
everything else. Goodman’s famous attack on resemblance (1972, 1976) is an
attack on resemblance as a theory of target (e.g. Greenberg, 2013; Walton, 1990).
However, Goodman’s theory is a non-starter, too, for it rejects the distinction
between content and target, reducing the theory of content to the theory of target.
The inability to distinguish between content and target is a source of confusion
in the theory of depiction. It suggests that the main problem is not that there is
no unified theory of depiction; maybe no such theory is possible. The problem is
that we have many theories that pick out different topics.
We are now in a position to characterize the iconic reference. The relationship
between an image, content and target is two-dimensional. The general idea is
that an image (in respect of the circumstances of evaluation) denotes a target
and the properties of the carrier exemplify the properties of content. In turn, the
properties of content (in respect of the iconic convention) are used to reidentify
(in terms of recognition-based identification) the referent. The content indicates
the target by means of reidentifying the referent. Depending on whether the
 What is an Image? 125

referent matches the target or not, the image is either accurate or inaccurate. Let
me call it the two-dimensional model of iconic reference (2-D for short). It is
represented in Figure 6.1.
Let us start with the concept of an image. In the 2-D model, it is the carrier of
the representational function. The carrier of a representation can be a physical
object, for example, a picture or a diagram, and a psychical one, for example,
a mental image. For instance, the carrier of the portrait of Duke Wellington is
the canvas covered with pigments, and the carrier of the mental image of Duke
Wellington is the mental representation that underlies the experience of the
Duke. The main question of the theory of iconic reference is how the properties
of the carrier, such as pigments on the picture’s surface, can represent something
outside of the carrier. In other words, what is it that turns an image into a
representation of something else?
Drawing on Goodman (1976), I distinguish between denotation and
exemplification as two kinds of reference relations. According to the 2-D model,
images denote the target and exemplify the content. Denotation is a dyadic relation
between a representation and something it stands for.3 Exemplification runs in
the opposite direction. It refers back from the selected properties of the object
to its representation. Additionally, it requires instantiation. A representation has
to have the exemplified property. For instance, a water sample exemplifies the
water quality, for it instantiates it. Exemplification is instantiation plus reference.
However, not every instantiated property is exemplified. A water sample
instantiates the property of being taken on a certain date. Yet, the sample does

Figure 6.1 The 2-D model of iconic reference representing the relationship between
an image, its content, the referent and the target. An image denotes a target in respect
of the circumstances of evaluation and exemplifies the content, which reidentifies the
referent in respect of iconic convention. The content indicates the target by identifying
the referent. © Piotr Kozak.
126 Thinking in Images

not exemplify the date. Depending on the context, only selected properties are
exemplified.
The role of a sample is to highlight the selected properties of the object. To
do so, the highlighted properties of the sample have to be salient. A bald man
wearing a hat is not a good sample of baldness. A sample has to be filtered (Elgin,
1983, 1996b; Goodman, 1984; Goodman and Elgin, 1988).
Let me explain the nature of iconic reference, starting with iconic denotation.
There are two general approaches to define it (Abell and Bantinaki, 2010). On
causal models, images denote their objects in virtue of being part of a causal chain
between the representation and the represented object. In intention-based models,
images denote in virtue of the author’s intentions which aim at specific objects.
The intention-based approach works well for paintings and drawings. Yet,
it hardly applies to photographs, for they are based on causal rather than
intentional relations. Causal models do not apply to fictional representations.
There is no causal relation between a unicorn and its picture, but there are
pictures of unicorns.
To find a way out, we need to make two remarks. First, these two approaches
are often conflated. Let us recall Kripke’s causal theory of proper names (1980). In
order to identify the denotation of a proper name, you have to fix the denotation
in the act of the initial baptism. However, the initial baptism is not part of the
causal chain, which is fixed intentionally. The causal chain is the way we identify
the intention-based act of baptism.
The general idea of the intention-based approach is that intention carries
a teleological function. A intends B in the sense that B is a goal of A. Causal
relations can be easily wired into such teleological functions. For instance, some
causal chains can be picked out by carrying some teleological function in a
biological system (Cummins, 1996).
In the case of photographs, certain causal chains can be picked out by
distinguishing the teleological function of the photographs. If a picture is
distorted due to a lens defect, the lens defect is part of the causal chain, but it is
not the target. We exclude this causal chain from the reference relation because
the photograph has not been designed to depict lens defects.
Second, intentions are not transparent. It is not the case that, when producing
an image, we have a clear view of our intentions, as if some transparent mental
events guide our actions. Rather, we infer our intentions by judging the final
product of our intentional act. Architectural sketches well illustrate this. If I start
sketching a building, I do not have to know what I want to draw. I am searching
for it.
 What is an Image? 127

Moreover, depending on the knowledge of accompanying causal factors, we

can change the interpretation of the target. For instance, if we learn some facts
about an author, we can reinterpret his paintings (Hopkins, 1998; Lopes, 1996;
Terrone, 2021a).
To reconcile the intention-based and causal theories, let us think of iconic
denotation in terms of the measurement theory. In measurement, the target
is the measured quantity. The measured quantity is the state of the logical or
physical space determined by different locations of information within this
space. For instance, 12 is localized within the logical space by its position in the
ordered numerical set. The objects in the physical space are localized within the
spatiotemporal frame of reference. The goal of the measurement is to identify
these states. Measurement refers to a measured quantity that it was designed to
identify.
By the same token, images denote their targets in terms of identifying some
states in the logical or physical space. An image denotes its target if it is aimed
at identifying these states. For instance, the inflation diagram denotes inflation
growth, and the black hole picture denotes the black hole, for they were designed
to depict them. In contrast, the Churchill-like trace left by an ant does not denote
Winston Churchill, for it was not designed to depict Winston Churchill.
The image target is independent of the content. Depending on the goal, the
same photograph can be used to depict a black hole or to test measurement
devices. In the same way, the same ruler can be used to represent straight lines
and distances. Interpretation of images is partly a matter of identifying these
goals.
Such interpretation is an ongoing process, for there are cases when an
image shifts its reference. We intend to portray Jane, but instead, we portray
Sarah, her twin sister. Before we knew that Jane had a twin sister, the picture
denoted Jane. After we learned that, the picture’s target is Sarah. By the same
token, a thermometer can be used to represent temperature. Yet, by learning
thermophysics, we can interpret it as representing pressure. Identifying the
target of a depiction is theory-laden and depends on our knowledge of the world.
At first glance, the measurement-based account of iconic denotation can
be seen as a refined version of the intention-based theory since identifying the
picture’s target can be based on identifying the intention of the picture’s producer.
However, there are significant differences. The measurement-based account
does not posit the existence of any mental event for identifying the relevant state.
A state can be identified by the use of an image in some epistemic practice. An
image is of something if it is used to identify the relevant state of space.
128 Thinking in Images

The source of confusion is that the concept of intention is nested within our
cognitive practice of explaining representational content. Ascribing intentions
can be seen as taking an intentional stance (Dennett, 1987) to explain the object
of representation. It can be useful to describe an image in terms of mental
events, such as intentions, to identify a targeted state. Yet, intentions are only a
method of identifying these states. The concept of intention is like the concept
of a meridian. It is used to identify some world properties but is not the same
object as a chair.
Does this mean that intentions are not real? This is a misplaced question. It
is like asking whether a meridian is real. These are certain ways of describing
the world, but they are not a description of the world. Depending on the level of
explanation, the intentional stance can be replaced by the design stance or the
physical stance. However, taking the design or physical stance does not falsify
the intentional stance. It is only a matter of the way we describe the world.
The measurement-based account can be reconciled with causal theories in
the following way. Identifying the target enables us to pick out relevant causal
factors. The black hole picture denotes the black hole, for it is part of a causal
chain whose initial element is the black hole. At the same time, the picture has
been designed to depict the black hole. Depicting it was a goal of the picture.
Therefore, irrelevant causal links, such as distortions caused by the defects of
measurement devices, are excluded from the denotation.
Yet, the existence of the causal relation is neither necessary nor sufficient for
denoting the target. The unicorn picture identifies a possible world inhabited by
unicorns and picks them as its target.4 Yet, there is no causal link between them.
However, this does not imply that causality plays no role in determining
iconic denotation. There are images, such as unicorn pictures, that denote
objects in some possible worlds, and images, such as black hole pictures, that
denote objects in the actual world. The key is to identify what possible world we
are in. The condition of identifying isomorphism between the properties of a
possible world and an image is too weak. A map of London can be isomorphic
with Mordor, but Mordor is not part of the actual world. Looking for causal
correlations seems to be a good criterion for such identification. Picking out
the causal relations between an image and the target indicates that the possible
world overlaps with the actual one.
Following Kaplan (1989), identifying the state of space and the possible
world we are in establishes the so-called circumstances of evaluation. It can
be defined as a function from the image to the possible world. By identifying
the possible world, an image can locate a target in a possible world w1 and be
 What is an Image? 129

targetless in w2. Compare a documentary and a feature movie. The target of the
first is located in the actual world; the target of the second is in a fictional world
(Terrone, 2021b).
To sum up, we arrive at the following definition of iconic denotation.

Iconic denotation: Image I denotes target T, iff (i) I is designed to represent T;

whereby (ii) I identifies a targeted state of a physical or logical space; in relation
to (iii) the circumstances of evaluation.

Let us turn to iconic content. In the previous chapters, I introduced the

approximate definition of diagram content. It has been described as representing
the construction rules allowing one to get to the searched object. The same
definition can be extended to the content of images. An image exemplifies the
iconic content to reidentify the referent by following the iconic convention.

Iconic content: An image I exemplifies content C, iff (i) I instantiates and

represents C; (ii) C identifies the construction rules and (iii) construction
invariants; whereby (iv) C reidentifies the referent R in relation to (v) the iconic
convention.

The general idea is that iconic content exemplifies the construction rule, which
determines the parameters of some informational space necessary to identify
the referent. Iconic content shows the ways to localize some state of physical
or logical space. We can unpack this definition based on the concepts of
recognition-based identification and iconic convention.
Primarily, an image does not exemplify the properties of the target. This
restriction is necessary, for, without it, the 2-D model would be vulnerable
to objections of the following kind. For one thing, exemplification could not
ground the content, for not every property can be instantiated. A picture of my
mother represents her as happy, but the picture is not happy. Similarly, there is
no way to exemplify the properties of unobservable phenomena. An image of
the Higgs boson does not instantiate any property of the boson (e.g. Frigg and
Nguyen, 2020; Mößner, 2018). For another, the symbol ‘word’ refers to a word
and exemplifies the property of being a word. Yet, it is not a picture of a word
(Goodman, 1988).
We can easily avoid these objections, for an image exemplifies the properties
of constructions, not the constructed objects. The properties of construction
and the constructed objects should not be confused. Consequently, images can
exemplify unobservable phenomena, as they do not show them. They show how
to localize them in some physical or logical space.
130 Thinking in Images

However, not every carrier property is a property of the content. Only these
properties of the carrier that are highlighted by the construction parameters
are relevant. For instance, I can draw a square imprecisely because of my
lack of skills. Still, I recognize it as a picture of a square, for it represents the
construction rules of squares. The content highlights relevant properties of the
representational carrier.
Notably, iconic content does not represent objects. It represents the ways
of identifying objects. Interpretation of iconic content is based on identifying
construction parameters determined by the construction rules. These
parameters are the basic units of construction, their order and the relation
between them. Construction rules determine the permissible transformations of
these parameters. Based on the iconic convention, we can identify construction
invariants, that is, the properties that remain unchanged across different
constructions. Finding these invariants brings about the recognition-based
identification of the referent. The referent is a property, object, event or scenario
identified by the content.
The construction parameters of iconic content are representational units,
such as colours and lines, and relations between them. They have to be
distinguished from the registration of the representational content. For instance,
brush strokes on the picture’s surface register the colour, but the representational
unit is a colour, not a brush stroke. Finding how to register content is a matter
of representational technique. In the same way, it is a matter of measuring
technique to use mercury (instead of, e.g., alcohol) to represent the temperature
in a thermometer. However, the representational unit is the height of the mercury
column, not mercury.
Consider a picture of John. The target of representation is John. However, the
way it is represented, the composition of lines and colours, is the content of the
picture. The referent is the object identified by these lines and colours. Brush
strokes are the ways colours and lines are registered.
The referent of iconic content can be either a particular or general object. Let
us recall the distinction between token- and type-diagrams. A token-diagram
represents a particular token, such as a specific triangle. A type-diagram
represents a general object, such as a triangle as a such. Depending on the
identified construction invariants, every image can be taken to represent either
a particular or general object. However, every type-representation involves
representing a token (but not vice versa).5
Importantly, recognition-based identification is directly referential. It implies
that iconic content is identified directly in the picture. It is not inferred from the
 What is an Image? 131

picture. However, this does not mean that representations have a mysterious
feature of being intentional in terms of seeing a nonphysical Bildobjekt (Husserl,
2005) or that we are seeing-in the represented objects (Wollheim, 1987). These
objects are artefacts of the measurement operations. How should we understand
this?
Consider the measurement operation. Saying that the term 20°C indicates the
particular temperature is not saying that we are intentionally directed to some
abstract number. This number is a construction property by which we localize
some state of the physical space determining its relations to other states, such as
19°C and 21°C (Armstrong, 1989; Swoyer, 1987). Saying that the temperature
is 20°C means that this number localizes some particular state of the physical
space. In this sense, we can directly identify this state by this number.
Let us compare it with iconic denotation. It has been characterized in terms
of dyadic relation to the target. In contrast, iconic content is not a relation to
some abstract referent. It is a way of localizing the referent. Localizing a referent
means finding its location in some physical or logical space, in the same way as
the number 20°C directly localizes some particular temperature value.
Images appear intentional, for we use intentional vocabulary to describe them.
For example, I can say that I see my mother in the picture. However, saying that I
can see my mother in a picture is not saying that I can see an abstract idea of my
mother. In the same way, saying that the temperature is 20°C is not saying that
the temperature has some numerical content. Iconic content, just as 20°C, is the
way to localize a certain state of the world.
Exemplifying content is insufficient to recognize the referent, for the
representational carrier can instantiate many different properties. Content
requires interpretation following the iconic convention. The goal of
interpretation is to select the properties that contribute to the recognition of
the target. Interpretation highlights these properties by identifying construction
parameters.
Interpretation of content is ruled by the Davidsonian-like principle of charity.
The principle holds that selected construction properties are exemplified to bring
us closer to the target. In other words, we presuppose that the goal of content is
to identify the designed target.
However, the nature of the principle of charity is heuristic. We can take
different interpretational stances and shift the target. For instance, we can
abstract away from the author’s intentions and study the social factors that are
causally manifested in a picture. In this sense, the picture’s interpretation is free
of the author.
132 Thinking in Images

Thus, we arrive at the following definition of the 2-D model of iconic reference.

2-D model of iconic reference: image I refers to target T, iff it (i) I iconically
denotes T; and (ii) I exemplifies iconic content C; and (iii) C identifies referent
R; and (iv) C indicates T through identifying R.

Importantly, matching the referent and target cannot be based on comparing

features of the target and referent; for then, we would have to reintroduce
descriptive content. Let us think about this matching in terms of the indexical
relation between the content and the target. The content indicates the target by
identifying referent in the same way as indexicals ‘here’ and ‘now’ indicate the
properties of space and time. How should we understand such a relation?
Drawing on van Fraassen (2008), let us compare this to identifying the
reference of the measured value. Saying that the thermometer shows 20°C is not
enough to say that there is a measurement relation. It could be simply coincidental
that the thermometer shows the number 20. There must be some direct link
between the measure and measurand. However, saying that the measured value
of 20°C matches the temperature is not holding that there are some similarities
between the number 20 and the temperature. To understand this proposition,
we have to know how to localize this value within the physical or logical space.
It is non-propositional knowledge of how to identify the property of being 20°C
within the measured space.
To locate an object in the logical or physical space is to determine which of
the possibilities it realizes. It is finding a particular localization in an array of
possibilities this space marks out. To locate the value of 20°C within the physical
space is to find the particular localization in an array of possible states of the
world. The result of 20°C indicates this localization.
Let us think about content in terms of representing a measure by means of
which we indicate a state of some space by identifying the parameters of some
search procedures. The content highlights certain properties to localize the
target. For instance, a portrait of John highlights certain properties by means
of which we identify John. These properties can be simple, as a composition of
lines and colours, and complex, such as being bald. The properties of the content
are like measurement parameters by means of which we identify the measured
object.
Let us consider a black hole picture. Based on our knowledge of the world, we
can localize a target of the black hole picture. It can be represented as a particular
state of the physical space represented by a set of possible states of affairs. The
goal of iconic denotation is to locate this set. Knowing the content and iconic
 What is an Image? 133

convention of the black hole picture, we can identify the referent. Recognizing
that the target matches the referent is knowing that they both localize the same
state in the physical space.
Maps illustrate this point clearly. The target of a map is a state of some
physical space, for example, London streets. The content of a map, for example,
the composition of directions and points, identifies the ways of localizing these
streets on the map. Knowing that the map preserves spatial relations, we localize
a street layout. This is a referent. However, to use the map to get about London,
we must locate ourselves on the map, that is, we must recognize that the map’s
street layout localization indicates the same localization in the physical space,
that is, London. We use the referent to identify the properties of the target. A
map is correct if the referent and target mark the same localizations.
Importantly, in contrast to the identification of a target, identification of the
referent is not relativized to possible worlds. Consider a cross-ratio invariant. A
geometric projection preserves cross-ratios in every possible world. No matter
what kind of space is projected onto the plane, whether we are in the Euclidean
3-D space or in the flat world, the cross-ratio is preserved in the projection. By
the same token, the triangle invariants, such as having three sides, are identical
in all possible worlds. The idea is that a referent localizes the same objects across
all possible worlds.
Consequently, depending on the kind of reference we pick out, an image
refers to the represented object rigidly or non-rigidly. From the perspective
of content, an image behaves like a rigid designator. From the perspective of
denotation, an image is a non-rigid representation. For instance, in a possible
world w1 a map of London denotes London, while in a possible world w2 the
same map denotes Mordor. However, in both worlds, the content of the map
is the same. Iconic content rigidly identifies the referent in the same way as the
standard metre identifies the same distance in all possible worlds.

Correctness conditions

The distinction between the iconic content and target expresses two kinds of
image correctness standards. In general, these standards are accuracy conditions.
Accuracy is a matter of degree. An image can be more or less accurate. However,
accuracy is a combination of two different kinds of standards. For one thing, an
image can be a true or untrue representation of the target. For another, iconic
content can be precise or imprecise. An image can be true but imprecise, and
134 Thinking in Images

Figure 6.2 The relation between trueness, precision and accuracy conditions. A
representation is accurate iff it is both precise and true. © Piotr Kozak.

precise but untrue. It is accurate iff it is both true and precise. The relation
between the concepts of trueness, precision and accuracy is represented in
Figure 6.2.
The concepts of trueness, precision and accuracy come from the
measurement theory. Trueness represents how close a measurement is to a true
value. Precision is the closeness of the measurements to each other (ISO, 1994).
We can apply these concepts to the analysis of iconic reference. First, the
general idea of the precision condition is that measurement results should be
unambiguous to identify the measured quantity. By the same token, the properties
of iconic content should be unambiguous to identify the referent. It implies that
the construction invariants should identify an unambiguous set of properties
that enables us to recognize the represented object. For instance, a photograph is
sharp if it identifies the properties that make us able to recognize the represented
object. Otherwise, it is unsharp. The inflation diagram is precise if it displays the
data that allows localizing inflation growth. Otherwise, it is imprecise.
The precision conditions are relativized to the goal of the representation
and the cognitive skills of the interpreter, which both build the context of
understanding an image. An image can be too little or too precise, depending on
 What is an Image? 135

the goal. For instance, a map has to be precise enough to identify localizations.
It cannot be too precise, for then it would overlap with the territory. Depending
on the cognitive skills of the interpreter, images can be understandable or not. A
mathematical graph is understandable only if its interpreter has the mathematical
competence to understand it.6
Second, a measurement is true if it matches the value of the quantity it was
designed to measure. By the same token, an image is true if the referent matches
the target. For instance, a picture of a square is true if it was designed to represent
squares. It is false if otherwise. I am not imprecise if I intend to draw a square but
draw a circle instead. I am wrong.
Importantly, we should not confuse the trueness of images with the truth
of propositions. The concept of trueness should be taken in metaphysically
neutral terms. We do not have to interpret it as presupposing the existence of
some true values (Giere, 2006; Swoyer, 1987). Instead, we can think of it in
terms of robustness conditions, that is, the concepts that introduce coherence
and consistency to measurement outcomes (Chang, 2004; Tal, 2013, 2017; Teller,
2013, 2018). For instance, the standard metre is not true in terms of representing
the true nature of what one metre is. It determines what one metre is. By the
same token, a painting can set standards of what beauty is. It does not imply,
however, that some abstract object exists called ideal beauty.
Yet, although the measurement-theoretic concept of trueness is distinct from
the concept of the truth of propositions, they are linked. Let me illustrate this
with the concept of a sample. A sample is true if it exemplifies the measured
value. For instance, the water sample is true if its quality is close to the water
quality. However, it is not guaranteed that the sample will represent it. The
sample can be badly taken, or the distribution of water quality values can be
abnormal.
In contrast, the truth of propositions requires that we can distinguish the
atomic formula of a semantical system and provide its translation into a
metalanguage that satisfies a T-schema. For instance, the proposition snow is
white is true for it satisfies the T-schema such that ‘snow is white’ is true iff
snow is white. By using the T-schema, we can determine the truth-conditions of
compound formulas. For instance, a compound proposition of the form A and
B is true iff both A and B are true. This implies that propositions have a logical
form that provides correctness conditions for such translation.
Samples cannot satisfy these conditions. They can be badly taken, but they are
not true or false in terms of the truth of propositions. No atomic formulas can be
distinguished, for any sample represents many properties at once. Samples have
136 Thinking in Images

no logical form. A sample of water quality does not logically imply water quality.
It instantiates water quality.
However, a sample can be a basis of a proposition in the following sense. The
water sample localizes the water quality property. It provides us with information
about the quality of water and can be a source of the proposition that water
is contaminated. The sample works as a justification for this proposition, too.
If someone asks why I hold that the water is contaminated, I can point to the
sample. Yet, this does not imply that I cannot be wrong. The sample can be badly
taken, and the proposition based on the sample can be wrong.
By the same token, an image is not true or false in terms of being a true or false
description of a target. Images can be properly or badly taken. At the same time,
images can be true in terms of the measurement theory if the referent identified by
the properties of the content is close to a target. An inflation diagram is true if it is
close to the true value of inflation growth. A conventional picture of a dragon is true
if it is close to the stereotypical depictions of dragons. An image of a square is true
if its properties are the same as those identified by the square definition. An untrue
picture is such that the referent localized by the content is far from the target.
To conclude, we get different correctness conditions depending on the kind
of iconic reference.

Precision condition: image I is precise iff the I’s content has such properties that
allow one to identify the referent unambiguously.

Trueness condition: image I is true iff the I’s referent is close to its target.

Accuracy conditions: image I is accurate iff I is both precise and true.

Two remarks are in order. First, the iconic reference and correctness conditions
are defined in non-mentalistic terms. They do not refer to the mental states of
a perceiver or a phenomenology of pictorial experience. However, it does not
imply that mental states are irrelevant to understanding the meaning of images.
No one holds that. It means that a theory of how we grasp meaning does not
overlap with the theory of meaning.
Second, the 2-D model heavily depends on our understanding of reference,
exemplification and indexicality concepts. It cannot be otherwise, for this model
is intended to be part of a general theory of meaning. The implication is that
the 2-D model can be incorporated into the general framework of the theory of
meaning. That being said, not all model concepts can be sufficiently explicated
since this would assume that we have to clarify other concepts of the theory of
meaning. No book is long enough to do that.
 What is an Image? 137

The measurement-theoretic account of images

So far, the measurement-theoretic framework has been applied to highlight the

measurement-like properties of images. This may suggest that the comparison
between measurement and images is allegorical only. However, the main
idea is that images are not like measurement devices. They are measurement
devices. Strictly speaking, images are a kind of measurement instrument that
shows how a measured object would look in some measurement set-up by
representing the ways of identifying this object. Let me start by unpacking
basic intuitions.

Image: a kind of measurement device that shows how a measured object would
look in some measurement set-up by representing the ways of identifying this
object.

First, there is a long-standing tradition of interpreting images in terms of

measuring devices. For instance, in the nineteenth century, photographs were
hoped to be scientific instruments comparable to telescopes (Maynard, 2017).
Second, we have good reasons to believe there is a strict analogy between
measurement and semantics. In general, measurement involves exploiting a
correspondence between a phenomenon and certain abstracta so that features
of the abstracta represent features of the phenomenon. Assigning semantic
values is like setting up a measurement system (Ball, 2018; Dresner, 2002).
Moreover, it is believed that the measurement theory offers a unified account
of the interpretation of language, action and mind (e.g. Dennett, 1987, 1991b;
Dresner, 2006, 2010, 2014; Marcus, 1990; Matthews, 2007, 2011; Swoyer, 1987).
Third, measurement is not reducible to the numerical domain. Generally,
measurements are about imposing order on a certain manifold (e.g. Finkelstein,
2003; Narens, 1985). Finding whether a couch fits into a doorframe does not
have to involve assigning any number to the couch.
What is a measurement? In the most general sense, a measurement consists
in a systematic application of some standard to some manifold to localize a
measured value. Primarily, it involves laying out what the measured quantity or
category is. Next, it provides a way of a systematic mapping of the measured value
or category onto the representation. Finally, it determines what must be done to
successfully carry out the measurement (Bradburn et al., 2017; Cartwright and
Runhardt, 2014; Chang and Cartwright, 2014).7
Every measurement is theory-laden. We need a theory or a model to
understand what is being measured. Every act of measurement depends directly
138 Thinking in Images

or indirectly on the outcomes of other measurements. For instance, measuring

air temperature depends on the outcomes of the measurement of air pressure.
Measurement is selective. Only selected properties are represented. It is
perspectival. It does not show the measured value from some absolute point
of view but in a measurement set-up. The measurement result is relativized
to the measurement system. Both the Kelvin scale and Celsius scale measure
temperature. However, the Kelvin scale represents thermodynamic temperature,
while the Celsius scale represents the intervals between temperature values (van
Fraassen, 2008).
Measurement involves applying some well-grounded metric, a function
that assigns distances to the locations of the measured values. Applying the
metric often involves setting up a scale that allows comparing the measured
values. For instance, the Celsius scale allows comparing intervals between
temperatures. However, not every measurement involves applying a scale
and comparing measured values. Consider Neurath’s Ballung concepts, such
as race, social exclusion, well-being or quality of life. Such concepts are too
multifaceted to be measured by a single metric without loss of meaning. They
must be represented by a matrix of indices or several measures. Moreover,
some categories are sui generis. The feeling you had the day your child was
born is incomparable to any other feeling. The sublimity of the starry sky
above me is not scalable.
I hold that the art of depiction is an art of measuring. Just like measurements,
images are systematic mappings of the depicted state onto a representation. They
are theory-laden. The acts of producing and interpreting images are embedded
within our knowledge of the world. Images are selective and perspectival. They
represent only selected information taken from some particular perspective.
This should not be taken as a trivial claim that images are an effect of using
measuring devices such as rulers. A map is based on measurements in the same
way as measuring velocity is based on measuring time and distance. In short,
measurement involves measurement.
Images are not representations of the measurement outcomes, either, as if
there were some measurement process separable from representing its outcomes.
If a thermometer indicates 20°C, the indication does not represent some prior
measurement process. It is part of the measurement.
Images are a kind of measurement devices, such as rulers (Boumans, 2005;
Morrison and Morgan, 1999). Let us consider the black hole picture. It is an
outcome of a measurement operation and a reliable indicator of the measured
values. Just like rulers, it localizes a phenomenon in some space.
 What is an Image? 139

Just like rulers and balances, images represent the ways of localizing the
referent. They represent the ways how to localize some states in a rule-governed
manner. They are not only applications of some measures. Images exemplify
these measures. Images represent the outcomes of measurements together with
the rules of construction, showing how to arrive at these outcomes.
Let us consider the inflation diagram. It does not tell us only about the
inflation growth value. It represents how to localize the inflation growth value
by mapping time onto the inflation rate. In the same way, a world map does not
tell us that London is localized at some geographic longitude and latitude only.
It also informs us how to get to London.
In contrast to rulers and balances, images can represent non-numerical and
qualitative properties. A colour sample shows you how to find the matching
colour. A depiction of John shows how to identify John. Images can represent
incomparable and unscalable phenomena. Think about the feeling you had the
day your child was born. I cannot describe it or ascribe it any number. Yet, I can
show you a way of how to arrive at it, for example, I can express it in music.
Thus, from a measurement-theoretic perspective, images play a twofold role.
On the one hand, they are the outcomes of some measuring procedures. For
instance, a portrait of John is an outcome of a measuring procedure performed
on the depicted situation, in the same way as 20°C is an outcome of measuring
temperature. An image localizes a depicted situation, just like 20°C identifies
a temperature value. On the other, images represent measures that determine
the ways we localize the measured values. Just as a ruler shows how to identify
a measured length, the portrait of John shows how to identify John. Images are
the tools by which we measure the world.8
What are the consequences of the measurement-theoretic approach for
understanding the nature of images? First, the question about the nature of
images is misleading. If we hold that images are a kind of measurement devices,
then we can characterize images only in functional terms. Let us think about
what connects mental, auditory and olfactory images. One can argue that the
use of the term ‘image’ in all these contexts is different. That suggests, however,
that there are essential properties of being an image, just like the atomic
number is the essential property of being a chemical element. According to the
measurement-theoretic approach, this assumption is false. Asking about the
common nature of mental, auditory and olfactory images is like asking about
the shared nature of rulers and balances. This is a badly posed question, for
there is no essential property of being a measurement device besides bearing a
measurement device function. We distinguish between different measurement
140 Thinking in Images

instruments depending on what and how is being measured. By the same token,
images can be characterized only functionally. They are the objects that are
necessary for some measurement operations. Mental, auditory and olfactory
images are different kinds of measurement instruments.
Second, according to the Traditional View, images are copies of the world.
In the measurement-theoretic framework, they are not copies, just as the ruler
is not a copy of some distance. Images may appear as if they were copies of
the world, just like a ruler may appear as if it were a copy of some distance.
However, just like rulers, images are measurement devices that represent the
measures to localize some state of the world and help us find our way in the
world.
Let us consider the portrait of John. According to the Traditional View,
the portrait is a copy of John. However, we can think about the portrait not in
terms of copying John but in terms of the ways of localizing John. According to
the measurement-theoretic approach, the portrait is a measure that we use to
identify John.
Perceiving the portrait of John does not imply that, to understand it, we need
to know what John looks like; in the same way, to understand what the standard
metre represents, we do not have to know what the length of one metre is.
Perceiving the portrait does not imply that when we meet John, we will see him
from exactly the same perspective. The portrait gives you a measure to localize
John, in the same way as the standard metre is a measure to localize the length
of one metre. This measure can be applied in different contexts, for it is not
characterized by copying reality but searching procedures.
Compare this to a map. On the one hand, it is an outcome of some
measurement procedures. On the other hand, it exemplifies a measure that helps
you find your bearings in space. We do not have to know what London looks like
to understand the map of London. Conversely, we use the map to learn where we
are. Moreover, to find out where we are in London, we do not have to pick out
the same perspective as that represented on the map. The ability to recognize its
spatial invariants and apply them in different contexts is crucial for using a map.
Does that imply that the portrait of John and a map of London cannot
misrepresent John and London? It does not. The portrait and the map, just
like any measurement, can be inaccurate. Is this not a vicious circle, however,
one could ask? To know whether a map is accurate, you need to know what it
represents. To know what is represented, you need an accurate map.
This problem can be generalized and applied to the 2-D model of iconic
reference. The ability to identify construction rules presupposes that we recognize
 What is an Image? 141

the final effect of the construction. Knowing how to reach a target presupposes
knowing where we want to get. At the same time, to recognize an object, we
have to identify the construction parameters. For instance, to recognize the
mountain seen in a painting, we must identify marks on the picture’s surface
that are used to represent it. However, to identify these marks, we must know
that they represent the mountain. Thus, the ability to construct an object has to
be taken together with the ability to recognize it.
However, the circle is not vicious. Consider the so-called coordination
problem in the measurement theory. It concerns the problem of coordinating
theoretical quantity terms with measuring procedures. The empirical adequacy
of the theory and the reliability of the measuring procedures presuppose each
other. To establish a theory, we have to test its predictions. However, this requires
a reliable method of measuring. This, in turn, involves having background
theoretical knowledge.
The traditional approach to the coordination problem holds that coordination
is accomplished by specifying definitions for some quantity terms. These
definitions are taken to be analytical and require no empirical testing. This solves
the problem of circularity but forces us to accept many metaphysically loaded
assumptions. The solution is particularly disturbing if we are not fans of the
analytic-synthetic distinction.
According to the coherentist approach to the problem (e.g. Chang, 2004; Tal,
2013; van Fraassen, 2008), we do not have to cut the circularity of coordination.
We can try to show that the circle is not vicious, for constructing a quantity
concept and standardizing its measurement are co-dependent and iterative.
Each iteration in the history of standardization respects and corrects existing
traditions. As van Fraassen (2008) argues, the coordination problem arises
when one adopts a foundationalist view and attempts to find a starting point for
coordination.
Let us consider the calibration function of measurement. It is inseparable
from any measurement activity, for every measurement is underdetermined
by instrument indications. The same indication may be taken as evidence for
multiple knowledge claims about the measured quantity, depending on which
background assumptions are used to interpret the indications. Calibration is
an activity of modelling different processes and testing their consequences for
mutual compatibility. However, comparing some standards is neither necessary
nor sufficient for successful calibration (Tal, 2017).
Let us apply the coherentist approach to the 2-D model of iconic reference.
An image is an effect of a continuing and co-dependent process of searching for
142 Thinking in Images

the best way to represent the target and trying to localize the target. It involves
ongoing attempts to match a referent to a target by searching for the most precise
way to identify the referent. Identifying the referent involves correcting our
background assumptions regarding the target.
This process is iterative and involves modifications of image construction
and representation target. It respects existing iconic conventions but is not
determined by them. Most importantly, it describes both the production and
interpretation of images.
Let us consider an architectural sketch. Sketching is a continuing process of
searching for the most precise expression of some architectural idea. At the same
time, it is a process of searching for the idea. When sketching, an architect has
only a rough measure of the target. He is localizing it by producing the content.
Next, think about the process of image interpretation. It is an ongoing activity
of trying to localize the target based on testing our assumptions in analysing
iconic content. At the same time, it is an activity of analysing iconic content based
on our background knowledge of the target. Broadly speaking, an architectural
sketch and image interpretation are a kind of calibration process of modelling
the idea and testing the consequences of the model.
To sum up, it seems productive to think of images as a kind of measurement
devices. This solves some old problems of depiction, such as the problem of
resemblance. On the other hand, it introduces some new problems, such as the
one of coordination. The advantage of the measurement-theoretic approach is
that it can be hoped that these problems can be solved within a more general
theoretical framework.

Application: Impossible images

In the following section, I show how the measurement-theoretic account of

images can be applied to impossible images. I chose this example for two reasons.
First, if it is held that we can see impossibilities in pictures, then a full-fledged
theory of depiction should be able to explain how to depict impossibilities.
The Traditional View does not make any room for impossible images: since
impossible objects cannot exist. Second, Sorensen (2002) offered a prize for a
depiction of a logical impossibility. I am going to show what counts as a sensible
response to his challenge.
Before going further, some restrictions are needed (Mortensen, 2010;
Mortensen et al., 2013; Sorensen, 2002). First, by impossible images, I understand
 What is an Image? 143

depictions of logical impossibilities (Elpidorou, 2016; Kulpa, 1987; Priest,

1999). Depictions of nomological impossibilities do not count, for we can have
depictions of dragons and fairies. Ambiguous figures, such as the Necker cube,
do not count. They entail different but consistent interpretations of logically
consistent objects.
Second, by impossible images, I understand depictions of logical
impossibilities open to inspection. Using a depiction of a dot as a picture of
a logical impossibility seen from a long distance does not count. Inaccurately
drawn geometrical figures, or figures with unnoticeable contradictory features,
are not impossible images, either.
Third, by impossible images, I understand consistent depictions of
inconsistent objects. Thus, if some consistent object is depicted from such an
angle that it looks inconsistent, the depiction does not count as an impossible
image. Similarly, we can have a visual illusion of a consistent object whose visual
interpretation is inconsistent, such as the Waterfall Illusion (Crane, 1988), which
is an illusion of the movement of stationary objects. Such inconsistencies are
common. Yet, they do not count as impossible images.
With these clarifications in mind, I discuss two examples of impossible
images: the Reutersvaard-Penrose triangle and the impossible fork. I argue that
they represent two kinds of constructions. They either represent non-orientable
objects or fail to construct an object.
First, let us consider the Reutersvaard-Penrose triangle (Figure 6.3).
The Reutersvaard-Penrose triangle9 depicts an impossible object, for, in order
to exist, such an object would have to violate the rules of Euclidean geometry.
For example, the bottom bar of the triangle is represented as being located to the
front and the back of the topmost point of the triangle. If such objects cannot

Figure 6.3 Reutersvaard-Penrose triangle. Source: Wikimedia Commons.

144 Thinking in Images

exist, then how is it possible to represent them? That is the paradox of impossible
images.
Some philosophers and psychologists try to find a way out of this paradox
by adopting three argumentation strategies. First, it is common to interpret
impossible images as instantiations of perceptual illusions (Ernst, 1986; Gregory,
1966; Kulvicki, 2006a). For example, it is possible to build an object that looks
like the Reutersvaard-Penrose triangle from a particular perspective.
However, the concept of an illusion implies that something merely looks like
something that it is not. No such thing needs to be said about the Reutersvaard-
Penrose triangle. Even if there can be a consistent object that resembles the
Reutersvaard-Penrose triangle, this does not imply that the triangle represents a
consistent object. If I build a scale model that looks like the Eiffel Tower, it does
not imply that photographs of the Eiffel Tower represent its model.
Second, it can be held that the Reutersvaard-Penrose triangle is an example
of a figure whose parts are consistent when taken separately but inconsistent
when put together (e.g. Blumson, 2014; Budd, 2008; Kennedy, 1974; Voltolini,
2015). According to this argument, just as we cannot have inconsistent beliefs,
we cannot have inconsistent images. Yet, we can have an inconsistent set of
beliefs that can be divided into consistent parts. Priest’s (1997) short story
‘Sylvan’s Box’ is an example of such a case. According to the plot of the story, at
one point, the main character believes that the box is open; at another, just the
opposite. The story is an example of a more general phenomenon. Any fiction,
just as experience, is full of more or less noticeable contradictions. Similarly,
the Reutersvaard-Penrose triangle is inconsistent as a whole, but each part is
consistent. Covering any two angles of the triangle reveals that the remaining
one is consistent.
However, this analogy leaves something out. In Priest’s story, we cannot
simply take out parts of Sylvan’s beliefs and hold that they are consistent. Such
a move would turn any contradictory system into a non-contradictory one.
We hold that the system is inconsistent if the relations between the parts of
the system are inconsistent. Thus, we need to supplement the story by holding
that Sylvan can be unaware of the inconsistencies of his beliefs. Alternatively,
he can have some implicit beliefs that render the inconsistent story consistent.
No such thing applies to the Reutersvaard-Penrose triangle. It is neither a
depiction of three disassembled parts nor is there anything implicitly assumed
in the picture. We see three assembled parts that are together inconsistent.
Nothing is hidden or implicitly assumed. The impossible object is seen in the
picture.
 What is an Image? 145

Third, the moral of Priest’s story can be interpreted differently. Instead of

asking about the consistency of the story’s parts, we may ask what can render this
story consistent. In ‘Sylvan’s Box’, the storyteller is reporting Sylvan’s beliefs that
the box is empty and not empty. Priest does not believe that it is possible. But it
can be possible within the story in the same way as we can affirm contradictions
in some logical systems, such as paraconsistent ones.
The same can be said about the Reutersvaard-Penrose triangle. It is a
consistent representation of some consistent object in some non-Euclidean
space that looks like a three-dimensional Euclidean space (Francis, 2007). It
looks like an inconsistent representation, for we implicitly assume that we deal
with an ordinary Euclidean space, which is false. Alternatively, we can stay
within Euclidean geometry and interpret the triangle as a type of the so-called
occlusion paradox, which means that it can be rendered consistent by changing
one or more occlusions (Mortensen, 2010, 119). You get a consistent triangle
representation if you rotate any two corners of the triangle.
Discussing the details of both solutions will not get us far, however, for
they cannot tell the whole story. Most of all, they leave us with a gap in our
understanding of the following example of an impossible image. Let us consider
the impossible fork (Figure 6.4).

Figure 6.4 The impossible fork. Source: Wikimedia Commons.

146 Thinking in Images

The impossible fork depicts an object that must have two and three prongs
simultaneously to exist. This cannot be the case. Therefore, it is an impossible
image.
No consistent mathematical theory can describe the impossible fork
(Mortensen, 2010, 135 ff.). It is not a type of an occlusion paradox, either. Changing
the occlusion of any part(s) of the figure does not make it consistent. According
to Mortensen, the only way out is to acknowledge the existence of paraconsistent
geometrical systems that can include objects such as impossible forks.
However, the price is that we have to acknowledge the existence of
contradictory objects in our ontological landscapes. Consequently, we have to
adopt some sort of dialetheism, which is a belief that there are true contradictions
(Priest, 1999), which is not a conclusion everyone can happily accept.
Dialetheism might or might not be true, but we can do without it. According
to the measurement-theoretic account, images refer to their objects by
exemplifying how these objects can be constructed. The general idea is that there
can be no impossible objects, but there can be ‘indeterminable’ constructions.
Constructions have been characterized as procedures of arriving at a target by
means of determining the parameters of some space. Arriving at a target means
localizing an object in this space. However, there are such constructions whose
parameters cannot determine the properties of some space to localize the target.
These are indeterminable constructions.10
There are two classes of indeterminable constructions. We can compare
them to attempts to measure unmeasurable properties and attempts to measure
with the wrong tool, such as measuring length with gas. The first class refers to
the constructions that aim at the spaces whose properties are indeterminable.
Consider the Reutersvaard-Penrose triangle. It represents a construction that
does not determine the properties of the triangle, for the bottom bar of the
triangle is located to the front and the back of the topmost point of the triangle.
However, this does not imply that it represents an impossible object. The
Reutersvaard-Penrose triangle can be interpreted as representing a construction
that cannot determine the kind of space it is in. Alternatively, it can represent
a construction in some non-orientable space, for example, a four-loop Möbius
strip. Non-orientable spaces do not determine such properties as orientation.
They may appear inconsistent if we confuse the concepts of non-orientability
and inconsistency. In this perspective, the Reutersvaard-Penrose triangle only
appears as if it represented impossibilities.
The second class of indeterminable constructions refers to the constructions
that fail to determine the properties of some space. These are abortive
 What is an Image? 147

constructions. They cannot reach a target since the construction parameters are
such that they localize an inconsistent object. Recall the operation of dividing
by 0. It localizes a set of inconsistent values. Constructing a triangle from the
line segments such that each side of the triangle is bigger than the sum of
the other two sides is abortive, for the operation cannot construct a triangle.
However, the fact that some constructions are abortive does not imply that they
do not exist.
The proper class of impossible images refers to such images that exemplify
abortive constructions. The impossible fork represents the rules of construction
that cannot reach the target. It is only one of the many abortive constructions.11
Moreover, since we distinguish between the properties of objects and the
properties of construction, we can keep the intuition that impossible objects
cannot exist, but there can be representations of impossible objects. Impossible
images represent constructions without targets. Consider a round square. Can it
exist? Obviously not. But there can be an image of a round square. Why is this so?
Metaphysical impossibilities are logical contradictions. The concept of a
round square is clearly contradictory. If so, according to the Duns Scotus Law, if
p & not-p, then q, everything can be an instantiation of this concept. However,
this does not seem right, for it contradicts the intuition that there can be better
or worse representations of non-existent objects. Compare an image of a shape
representing a cross between a square and a circle (the so-called squircle) and an
image of a horse. Granted, they are both inaccurate depictions of a round square.
However, the squircle appears less inaccurate than a horse picture. The squircle
represents an inconsistent measure to localize the round square, but it is still
better than the horse measure. By the same token, a ruler made of elastic material
is a bad measure, but it is still better than a ruler made of gas. In both cases, we
will get inconsistent results; yet in the first case, they will be less inconsistent.
To sum up, in the measurement-theoretic framework, we do not have to
introduce inconsistent geometries to explain impossible images. Moreover,
we can deliver a unificatory explanation of two kinds of impossible images.
Impossible images are the kind of images that exemplify indeterminable
constructions.

Summary

In this chapter, I described the 2-D model of iconic reference. According to the
model, images have two-dimensional semantics. They denote their targets and
148 Thinking in Images

exemplify content that identifies the referent. In the 2-D model, images employ
two kinds of correctness conditions characterized in the measurement-theoretic
terms: trueness and precision. Images that are both true and precise are accurate.
I employed the measurement-theoretic framework to explain the properties
of iconic reference. I claimed that images are a kind of measurement devices.
Contrary to the Traditional View, images are not copies of the world. They are
ways to localize the properties of the world, comparable to rulers.
In the last section, I showed how this framework could be applied to explain
the phenomenon of impossible images. According to the measurement-theoretic
view, impossible images represent indeterminable constructions.
 7

Thinking with images

In the last chapter, I described images as a kind of measurement devices. I

argued that images represent the world in a twofold manner: by denoting their
targets and exemplifying the rules of construction that reidentify the referent.
If the referent matches the target, then an image is accurate. In contrast to the
Traditional View, images are not copies of the world. They are measurement
devices, such as rulers and balances, used to localize the states of the world.
In the following chapter, I demonstrate how the measurement-theory
account of images can help explicate the imagistic theory of thought. Primarily,
I explain how this account addresses three kinds of challenges to the imagistic
theory of thought. According to the Epistemological Challenge, the imagistic
theory of thought cannot provide a theory of knowledge, for images are not
truth-evaluable and lack logical form. According to the Semantical Challenge,
the imagistic theory of thought cannot provide a theory of content as the object
of imagistic thought is impossible to determine. According to the Metaphysical
Challenge, the imagistic theory of thought cannot explain the systematicity and
compositionality of thoughts.
Finally, I show how the measurement-theoretic account can be applied to
the problems of representational format and mental imagery. Particularly, I
demonstrate the metaphysical consequences for thinking about the structure of
mental representations and the nature of mental imagery.

Imagistic knowledge

The backbone of the Epistemological Challenge is Wittgenstein’s argument from

content indeterminacy and Frege-Davidson’s argument from lack of logical
form. Their common assumption is that they implicitly posit the Traditional
View. However, these arguments do not apply to the measurement-theoretic
account of images. Why is that so?
150 Thinking in Images

The general idea of the Traditional View is that images resemble the world.
From such a perspective, it is always possible to ask whether iconic content fits
the resembled reality. Thinking about imagistic content in terms of matching the
world suggests that images can be true or false. However, these assumptions are
questionable if we accept the conclusions of Wittgenstein’s and Frege-Davidson’s
arguments. There is no room for pictorial truth.
It may appear that the lack of truth values prevents images from being
part of a rational train of thought. However, we can accept the premise without
acknowledging the conclusion. We do not need truth for something to be part of
a rational train of thought. Models and idealizations are not true, yet they are a
vital part of knowledge (Cartwright, 1983; Elgin, 2017). A thermometer is not a
truth-bearer. However, its readings can be part of justified beliefs.
In the same way, images can do epistemic work without being true (in the
propositional sense). A geometric drawing justifies our beliefs that a mathematical
figure is constructible. A map justifies our belief that we can get from London
to New York (e.g. Gauker, 2020; Shepard and Cooper, 1982; Williamson, 2016).
Moreover, images can test our beliefs. If I believe that triangle sides can have
lengths such that each side is longer than the sum of the other two, I can change
my mind having been shown that such triangles cannot be constructed. If I hold
that London is west of New York, I can change my mind by looking at a map.
Does this mean that images provide a reliable source of knowledge? It depends
on our epistemic goals. Using measurement devices is context-sensitive. Rulers
are usually reliable measurement devices, but sometimes they fail to meet our
needs. A ruler can identify the length of line segments, yet it is useless when it
comes to measuring the distance between the Earth and the Sun. Usually, we
rely on images. Sometimes they are unreliable. However, this does not imply that
we can dismiss images as useless in our epistemic practice, just like we do not
dismiss rulers as useless.
Thus, images appear to provide some knowledge. The question is what
such imagistic knowledge can be. If images are not true, they cannot provide
propositional knowledge, for the latter involves a truth-evaluable description of
the world. By the same token, images cannot provide information about possible
states of affairs, for it would require them to be truth-bearers. Neither can they
provide explanations, for the latter are propositional and have logical form. The
explanation is shaped by argument and involves inferential structures. Images
are not ‘inferentially promiscuous’.
It may seem that we can respond to these doubts by biting the bullet and
distinguishing between propositional and non-propositional knowledge. One
 Thinking with Images 151

can hold that although images are not propositional, they can constitute a form
of non-propositional knowledge.
However, much depends on how the concept of non-propositional knowledge
is understood. It is most often argued that imagistic knowledge, contrary to
propositional knowledge, is non-conceptual in content and builds phenomenal
knowledge or knowledge-by-acquaintance (e.g. Gauker, 2011; Gregory, 2013;
Mößner, 2018). For instance, a memory image of a red rose gives us knowledge
of what rose-redness looks like (e.g. Tye, 2012).
This strategy, however, will not get us far. First, non-propositional knowledge
can have the same correctness conditions as propositional knowledge. Depicting
what a thief looks like can be put in terms of a set of propositions attributing
properties to the thief. For instance, depicting a thief as bald is correct under the
same conditions as my belief that the thief is bald. If I depict the thief as having
long hair, I depict him incorrectly in the same way as the proposition the thief
has long hair may be incorrect. That implies that if I know what the thief
looks like, I know that he is bald. In other words, if I know what x looks like, I
know that in such and such circumstances, x looks so and so (e.g. Stanley and
Williamson, 2001).
Second, we rightly expect images to inform us about the world. For
instance, if I take a picture of a thief, I expect to learn who the thief is, whether
he is bald or tall. I am not interested in how the thief is presented to my
mind. When scientists study the black hole picture, they do not want to learn
about the phenomenology of black holes, either. They want to know what
black holes are.
Thus, images should tell us how the world is and not only what it looks like.
If images could provide only phenomenal knowledge, then imagistic knowledge
would be useless from the point of view of our epistemic interests. Instead, we
rightly expect images to figure within the web of our beliefs and theories.
The Epistemological Challenge questions these expectations. In other
words, the role we attribute to images cannot be fulfilled. That is because we
hold that beliefs and theories are truth-evaluable; images are not. Images are
not propositional, yet we expect them to figure in propositional knowledge. The
problem is not that these expectations are not right. The question is how they
should be explained.
The lack of propositional knowledge can be compensated by an image’s
contribution to understanding. As it seems, images are a plausible source
of understanding without being a tool of explanation (e.g. de Regt, 2017;
Lipton, 2009, Mößner, 2018; Zagzebski, 2019). The black hole picture gives
152 Thinking in Images

us an insight into what black holes are. It does not provide a theoretical
explanation of why they are so. That being said, we need a clear account of
what understanding is.
Importantly, we need a view of understanding that is not reducible to
explanation. For instance, coherence-based accounts of understanding (e.g.
Elgin, 2017; Kosso, 2007; Kvanvig, 2018) define it as recognizing connections
between facts and as a skill of matching things to form a general schema.
However, recognizing coherence between facts is only another way of knowing
an explanation of these facts since the explanation is based on identifying factual
connections (Khalifa, 2017).
According to de Regt (2017), understanding is distinct from explanation. It
does not require true theories, either. It is a skill of making theories intelligible
but not necessarily true. A theory is intelligible if scientists can recognize some
characteristic consequences of the theory qualitatively. For instance, a kinetic
picture of gas as a collection of molecules in motion enables a qualitative
understanding of how gas behaves. By developing molecular models of gas,
one can arrive at intelligible kinetic theories of gas which are the sources of
constructing explanations of the gas phenomenon.
Although de Regt’s concept of understanding does not directly apply to
imagistic knowledge, it hints at how to think about the latter. According to the
measurement-theoretic account, images are measurement devices representing
the ways of identifying measured objects. The idea is that they enhance
understanding by providing recognition-based identification of the objects of
our beliefs and theories.
Recognizing objects is not a propositional kind of knowledge. It is not a
matter of knowing a description of an object, either. I can know a description of
an object without being able to recognize it. Recognition is the skill of localizing
the represented object in some space. Knowing how to recognize an object is a
non-propositional (procedural) kind of knowledge.
Let us compare this to the understanding of indexicals. According to Perry
(1979), understanding indexicals such as ‘here’ and ‘now’ is irreducible to
knowing a description of the space you are in. Believing that a meeting starts at
noon is irreducible to knowing that it is now. Knowing that to leave a wilderness,
you have to follow the Mt Tallac trail is different from knowing that the Mt Tallac
trail is here. Indexicals locate our beliefs in space and time.
By the same token, imagistic knowledge provides an understanding of the
objects of our beliefs and theories by localizing these objects in some logical or
physical space. What does it mean?
 Thinking with Images 153

Imagistic Knowledge: the skill of recognizing objects of our beliefs and theories
within the logical and physical space.

Let me illustrate this with the case of interpreting a map (van Fraassen, 2008). To
understand our position in space, we have to localize ourselves on the map. This
is not the kind of information we can get from the map’s description. Unless we
can find our location on the map, the map is useless. Even if a map indicates the
point ‘we are here’, we still have to determine our position. Based on the map, we
have to localize ourselves by identifying the parameters of the space. Moreover,
we need to know what moves are permissible by the map. For instance, we need
to know that if we turn left, we can get to point A but not B. To interpret the map,
we must identify how we can move.
The operation can be quite simple. It can be based on reading off the
information ‘Diagon Alley’ and finding the same inscription on the street. Next,
based on our knowledge of how streets are plotted on the map, we can find our
way from Diagon Alley to Knockturn Alley. Thus, a successful interpretation of
the map involves our ability to localize and orient ourselves in the space. What
does this teach us about the nature of imagistic knowledge?
According to the 2-D model, a map is a measurement device used to orient
ourselves in the logical or physical space. It exemplifies the rules of construction
and construction invariants. These invariants are the landmarks, such as street
names and the spatial relations between them. They enable us to identify spatial
localizations and directions. A successful interpretation of the map involves
identifying these spatial invariants and recognizing permissible moves so that
one can orient oneself in space. Consequently, a good map helps us to understand
our position in the physical or logical space.
By the same token, imagistic knowledge enhances the understanding of our
beliefs and theories by means of localizing them in the logical or physical space.
Such knowledge is irreducible to the set of propositions in the same way as
understanding a map is irreducible to knowing the map’s description. Imagistic
knowledge is the skill of recognizing the objects of our beliefs and theories
within the experience and the space of possible states. Let me illustrate this with
two examples.
Suppose you want to understand what a harmonic oscillator is. This involves
being able to point at a pendulum and hold ‘this is a harmonic oscillator’. The
pendulum is a model that identifies the invariant properties of the harmonic
oscillator by means of which you localize the harmonic oscillator within the
space set up by classical mechanics. The model does not explain the theory. It
154 Thinking in Images

localizes the states set up by the theory. It helps us to orient ourselves within the
theory, making it intelligible.
In the same fashion, the black hole picture makes the theory of general
relativity intelligible.1 The theory offers explanations of the origin and behaviour
of black holes. It provides a theory of what black holes are. To understand this
theory, you have to be able to localize the object of the theory in the physical
or logical space. You have to be able to point at the black hole picture and hold
‘that is a black hole’. The black hole picture enables recognizing the phenomenon
the theory describes. It helps you to orient yourself within the theory of general
relativity.
Knowing how to orient ourselves within the theory is a non-propositional
kind of knowledge. It consists in knowing how to localize the object within
some logical or physical space set up by the theory. Such information cannot be
deduced from the theory. The role of a theory is to explain a given phenomenon.
In contrast, images localize the object of the theory by enabling recognition-
based identification of this object.
By the same token, images identify the objects of our beliefs. Let us consider
a portrait of John. The picture identifies John, just as indexicals ‘here’ and ‘now’
localize an object in spacetime. The portrait exemplifies the construction rules
by means of which we identify John. Such information is irreducible to knowing
a description of John. I can recognize John without knowing a description of
John; I can know a description of John without any ability to recognize him in
the picture, too.
Images are irreplaceable in imagistic knowledge. Knowing how to recognize
objects is irreducible to knowing the descriptions of these objects and, therefore,
cannot be represented by any set of propositions. The picture of John can
misrepresent John, but it does not express any false proposition. I can recognize
John in the picture without any descriptive knowledge of who he is. The only
known medium that can represent imagistic knowledge is an image, for it
exemplifies the rules of construction by means of which we identify the depicted
object.
To illustrate this irreplaceable role of images, let us consider the mental
rotation tasks (Shepard and Cooper, 1982; Shepard and Metzler, 1971). The task
is to create a mental image of an object and rotate it in order to compare it to
the presented figure. Next, one has to decide whether both objects are the same.
Notice that such a decision cannot be made simply by comparing descriptions
of the object expressed in some language-like representation system (Pylyshyn,
1973). First, you need the skill of rotating the first object in the right way. There
 Thinking with Images 155

are infinitely many ways to rotate this object, but only one way leads to solving
the problem. The way of rotation cannot be inferred from the figure’s description.
If it could, it would require checking infinitely many ways of rotation which
leads directly to combinatorial explosion.
Second, we would still need additional information if we had two similar
descriptions of the figure. The task is to identify two objects as identical.
This information does not follow from having two similar descriptions. I can
know these descriptions without any ability to recognize that they refer to the
same object. I can know both the content of a map and a description of my
environment, but it does not follow that I can orient myself in the space. By the
same token, I can know a description of John, but I still need the information
that the depicted person is John (Kaplan, 1989).
Granted, the description-based explanation of how people solve mental
rotation tasks does not have to be wrong. The problem is that it is not sufficiently
informative to work as an explanation.
In the 2-D model, the way we solve mental rotation tasks is easier to
understand. Recognizing two objects as the same involves identifying the rules
of construction. Images exemplify the rules of construction by means of which
we recognize depicted objects. We recognize two figures as the same if we
identify the same construction invariants determined by these rules.
In the mental rotation tasks, subjects hold that they visualize the process of
the figure rotation. Suppose the measurement-theoretic account of images is
correct. In that case, the fact of visualizing this process results directly from the
epistemological properties of images. Only images can convey the information
that is needed in mental rotation tasks. The pictorial character of mental imagery
is not a contingent fact of human psychology but a necessary semantic fact.
Although imagistic knowledge is irreducible to propositional knowledge,
they are both entangled. First, localizing the phenomenon that is the subject of a
theory enables testing the theory. The epistemic role of images is to localize the
states within the space set up by the theory. If these states contradict each other,
it is a good reason for rejecting the theory.
Let us suppose that the black hole picture contradicts our predictions resulting
from the theory of general relativity. If a theory predicted that black holes were
square-shaped, but the black hole picture identified a crescent shape, it would be
a good reason to hold that the theory is wrong rather than poorly illustrated by
the picture.
Second, images can be a basis of propositional knowledge without having a
propositional character. According to the 2-D model, images do not describe
156 Thinking in Images

the world. They represent the ways we identify the states of the world. Images
represent how the world can be perceived. Consequently, images are not true or
false in the way propositions are. Images can accurately or inaccurately localize
the target. They are not expressing any truth. Propositions do that.
However, based on identified objects and properties, we can form beliefs and
theories. Images are the measurement devices we use to inform ourselves about
the world. A measurement for its own sake is a fruitless endeavour.
Let us consider the case of a ruler. The ruler’s length is neither true nor
false. It represents the way we isolate a spatial magnitude. At the same time, the
indication of the ruler can provide a basis for describing distance. Based on that
indication, I can hold that a line segment is one metre long and so on. We use a
ruler to describe its target.
Similarly, we use images to describe the world. Pointing at the black hole
picture, I can hold that the black hole emission ring is crescent-shaped. The
picture is used to describe the black hole by isolating its spatial features.
Images enable recognition-based identification of objects. However, images
cannot tell us what these objects are. To interpret an image, we need propositional
knowledge.
Compare this to the way we interpret measurement indications. There is
a difference between measurement indications, such as the numerals on the
thermometer display, and measurement outcomes. The latter are knowledge
claims that associate the measured values with the object being measured and
are inferred from one or more indications along with relevant background
knowledge (Tal, 2017). The reading of the thermometer indicates some physical
value. However, it is the role of a theory to interpret this value. Without a theory,
we can identify objects but cannot know what they are.
Similarly, using images involves some background knowledge necessary to
interpret their meaning. For instance, the black hole picture provides perceptual
recognition of the black hole. Yet, we need a physical theory to know that the
picture represents a black hole.
With these distinctions in mind, we are now in a position to address
the problem of phenomenological content. Let us recall Gregory’s idea
of distinctively sensory content. According to Gregory (2013), pictures
represent the ways something is like. They represent qualitative aspects of
experience. For instance, Cezanne’s Mont-Saint-Victoire paintings represent
distinctive ways in which the mountain appears to us. They represent the
colour depth and the way the light illuminates the mountain. How should
we explain this?
 Thinking with Images 157

The Traditional View appears to be unable to explain the phenomenological

nature of imagistic knowledge. First, the Traditional View requires comparing
two relata of the resemblance relation. This implies that we would need to
access two spatiotemporally different quale simultaneously, which cannot be
right. I cannot have simultaneous access to how Cezanne’s Mont-Saint-Victoire
paintings look and what it is like to see the mountain.
Second, it gives us no insight into how to determine the conditions of
correctness, for it would require access to the painter’s private experience.
Cezanne painted several versions of Mont-Saint-Victoire. Without knowing
what his experience was like, we cannot determine which of the paintings
represents the mountain correctly.
From the measurement-theoretic perspective, to understand the sensory
content of a painting, we need to be able to localize the represented quale in some
space. Let us think of it in terms of constructing a map. Suppose that every quale
represents a point on this map. Image properties determine the parameters of the
space that the map represents. They are the means by which we localize a particular
point on this map. Images are measurement devices we use to identify this point.
Consider a particular shade of green represented in Cezanne’s painting.
According to the 2-D model, it is not a copy of the quale experienced by
Cezanne. It is the construction parameter by which we identify the particular
shade of colour. The painting represents the way we localize this colour in our
experience. Cezanne’s painting exemplifies a measure we use to recognize this
colour in experience. For instance, we can point at some colour and hold ‘that is
the colour I have seen in Cezanne’s painting’. We see through Cezanne’s paintings
just like we identify length by using a ruler.
Importantly, the ability to identify a particular shade of the colour in the
painting is necessary to understand this colour. Consider two cases. First,
suppose that you know a description of the colour in the sense of knowing its
coordinates in the RGB system. To make this description intelligible, you need
to be able to localize it in some logical and physical space. Primarily, you need
to know what RGB stands for, and therefore you need to know, in advance, what
red, green and blue are. In other words, you need the skill of identifying these
colours.
Second, we can describe a colour shade as a particular wavelength of light
experienced by a subject. However, this wavelength of light is precisely this
shade of colour (Williamson, 2002). The ability to find this colour is part of
understanding this description. A colour sample in the painting does exactly
that. It localizes this description within the space of colours.
158 Thinking in Images

According to the 2-D model, Cezanne’s painting is a measurement device

by which we identify the world’s phenomenological properties. This does not
imply that it provides information on what these properties are. There is always
some background knowledge involved in the process of interpretation. For
instance, we have to know some facts about the author to know that the painting
represents Mont-Saint-Victoire. That is not the kind of information that we can
get from the painting. However, this is not the image’s purpose.
In a nutshell, images exemplify measures by which we identify properties,
objects and events. They enable us to understand the objects of our beliefs
and theories by means of localizing these objects in some logical and physical
space.

Iconic content

According to the Semantical Challenge, the imagistic theory of thought cannot

provide the theory of content. Images cannot determine the content of the
predicative function and individuate the object they predicate of.
The Semantical Challenge and the Traditional View are closely bound
together. According to the Traditional View, image A refers to object B in a
way that A resembles B. However, to determine that A resembles B, we have to
individuate the object that is being resembled. To individuate this object, we have
to determine what properties are depicted. Thus, the structure of resemblance
relation mirrors the structure of a predicative function.
The Semantical Challenge holds that predication requires a structure that
distinguishes between the object of predication and the property we predicate.
Propositions and language have such a structure. Images do not, for their content
is indeterminable, and they lack logical form.
In the 2-D model, the Semantical Challenge is misdirected. Primarily, the
content of images does not involve any predication. Iconic content is not a
description of some object. It represents the way of localizing this object.
Let us consider thermometer readings. If I hold that the air temperature
is 20°C, I do not hold that the temperature is in some dyadic relation to the
number 20. I do not hold that 20 resembles the temperature, either. The number
20 does not predicate of this temperature that it has the property of being even or
being twice as much as the temperature 10°C. The number 20 is a measurement
predicate that represents the way we identify a physical magnitude by finding its
place on a scale. This number localizes the particular state of the world and its
 Thinking with Images 159

relation to other states. It shows how this temperature can be localized in some
measurement set-up.
By the same token, an image of John being bald represents the way of
identifying John. Depicted properties are the information we use to localize
John. When we depict John as bald, the property of being bald works as the
measurement predicate used to identify John. It does not predicate anything of
John, as if there were some relation between these properties and their object
seen in the picture, that is, something we predicate of and a predication. Depicted
properties represent the way we see John, not the way John is.
Let us compare a depiction of John with the proposition John is bald. The
proposition individuates the object of predication, that is, John, and predicates
of him the property of being bald. The property of being bald is the property of
the predicated object.
In contrast, we do not hold that the marks on the picture’s surface are bald or
that John is made of paint. These marks are the properties of the representation,
not the properties of the represented object. According to the 2-D model,
the properties of representation are the properties of the construction. Thus,
representing John as bald identifies the ways of arriving at John. Mistaking
the properties of construction and the constructed object leads directly to the
pictorial fallacy.
In the measurement-theoretic perspective, images do not bear predicative
functions. However, images can be a basis of predication in the following way.
Suppose you measure some object with a ruler. The indication of the ruler,
such as ‘one metre’, does not predicate the property of being one metre of this
object. It identifies this property. By using a ruler you establish the length of the
object. However, once this value is found, we can predicate it of this object by the
proposition this object is one meter long.
By the same token, a portrait of John does not predicate anything of John.
It localizes John by determining the properties that serve to recognize him.
For instance, it represents the property of being bald which is a measurement
predicate used to identify John. Once you recognize John in the picture, you can
predicate of John that he is bald.
Language is essential to express predicative functions (e.g. Davidson,
2001; Devitt, 2006). It distinguishes between the object of predication and the
predicate. It distinguishes between different modal contexts. Images cannot do
that. However, their job is to identify objects, not to predicate of them.
Moreover, the role of language is to determine the kind of predicate we project
onto the predicated object. Consider an image of a red square. It identifies the
160 Thinking in Images

properties of being red and square. However, the image does not predicate of red
that it is square-like or of a square that it is red. Language does that. Only within
the linguistic categories can we distinguish between the meaningful statement
that a square is red and a meaningless expression that red is square-like.
In the measurement-theoretic account, language and images are semantically
linked. On the one hand, recognizing the iconic content is a prerequisite to
describing this content in language. For instance, if pointing at a picture of John,
I hold ‘it’s John’, I have to be able to recognize John in the picture. Recognizing
John is a skill that is not language-based. In contrast, it is necessary to form a
description of John.
On the other hand, language is essential for making images meaningful. I
can recognize John in the picture, but without language, I cannot represent the
thought that the picture represents John. By the same token, we need language
to interpret the measurement’s results. Thermometer readings are meaningless
unless we can express what they represent. Only language can do that.
Does this imply that the iconic content is indeterminable and needs language-
like representations to fix its meaning? It depends on how we understand this
question.
Let us make two preliminary remarks. First, to determine the content of
representation, we need a criterion by means of which we decide what the object
of representation is. Second, in the 2-D model, we distinguish between rigid
and non-rigid kinds of reference. Respectively to the context of evaluation, an
image denotes an object non-rigidly. The iconic content, however, works as a
rigid designator. It identifies the same referent in every possible world.
The general idea is that we can introduce the criterion only in the case of
non-rigid kinds of reference. It is always possible to ask about a criterion that
determines image denotation. For instance, I can hold that the target of the
portrait of John is John since the author intended to depict John. Here, the
intention is a criterion by means of which we identify the target respectively to
the context of evaluation. We use this criterion to determine the target of the
portrait. Moreover, propositional representations are essential here. Applying a
criterion requires that we have a set of beliefs that are a basis of background
knowledge that justifies the use of such a criterion.
The question about criterion, however, is senseless in the case of iconic
content. We cannot ask about the criterion that determines iconic content. This
type of content refers rigidly to the referent. It identifies the same referent in
every possible world regardless of the context of evaluation. Iconic content is
criterion-less and directly referential. Let me explain.
 Thinking with Images 161

According to the 2-D model, images exemplify the rules of construction by

means of which we identify the referent. Here, the question about criteria cannot
be brought up. Just like a ruler sets up the criteria for identifying distance, an
image sets up the criteria of referent identification.
Let us compare this to the Traditional View. If we say that pictorial content
resembles some object, we can always ask about the criterion of resemblance.
It is possible since we can distinguish between the copy and the copied object.
Moreover, the resemblance criterion appears to be primitive and basic. It
does not involve any dictionary or a syntactical structure. We directly see the
resemblance. Both assumptions, however, make the Traditional View vulnerable
to Wittgenstein’s indeterminacy argument.
In contrast, in the 2-D model, iconic content is not a copy of the target. It is
a representation of the ways by which the target is identified. Here, the question
about a criterion is meaningless, for we cannot distinguish between what is
being represented and how it is being represented. There is no place for a middle
term between the image and a referent, for the image exemplifies the referent.
The source of the confusion is that iconic content represents some rule-
governed procedures of arriving at some object but is not a description of this
object. Images exemplify the way this object can be taken.
To illustrate this, let us consider the following example. Suppose that I show
you how to construct a triangle. If I am drawing a triangle, I am not predicating
anything of this triangle. I am showing you how to arrive at it. This drawing has
certain properties, for example, the lengths of the line segments. These are the
properties of construction, allowing you to identify a triangle. The lengths of the
line segments are the properties of the searching procedure by means of which
you arrive at the desired object.
At the same time, the length of these line segments identifies the object’s
property. If I show you how to construct a triangle, I arrive at some object with
determined side lengths. The length of the triangle line segments is the final
effect of some construction procedures. We cannot separate these properties
and ask about the criterion by which we determine representational content.
The way the object can be depicted is the content of the depiction. The image
demonstrates that if you take these line segments and connect them properly,
you will arrive at this object.
Importantly, we should not confuse the properties of construction with the
properties of the constructed object. The triangle construction identifies the
length of the searched object’s line segments, yet, it does not describe this object.
By the same token, the length of the ruler represents the way of identifying
162 Thinking in Images

the measured magnitude; yet, the length of the ruler is not the property of the
measured length. The ruler represents the way of localizing the object’s length.
Let us compare this with a description of a triangle. If I describe what a triangle
is, I predicate some properties of the mathematical object. The description can
be wrong, depending on whether the properties identified by the description
match the object’s properties. You need to find the described object to determine
whether a description is correct.
In contrast, a triangle diagram does not have to be compared to some abstract
mathematical object. The image tells you that if you take such and such lengths
of the line segments and arrange them properly, you will arrive at the object
represented by the image. With a triangle diagram, you do not say that now
you can look for triangles to find out what they are. You do not have to look for
anything that can match the image. The image is the thing you are looking for.
Compare this to the portrait of John. Suppose that you want to know who
John is. In this case, I can point at his portrait. The portrait represents a way of
identifying John. However, if you are shown a portrait of John, you do not say
that now you know how to find out who John is. The portrait is the thing you
are looking for. The portrait exemplifies the standard you use to identify the
referent. It does not require a criterion you use to identify the content, for its
iconic content determines the criterion of referent identification. The portrait of
John exemplifies the criterion by which you identify John, in the same way as a
triangle diagram exemplifies the criterion by which you identify triangles.
By the same token, a ruler exemplifies a standard used to identify spatial
distances. I can point at the ruler’s length if you ask me what one metre is.
However, it is nonsensical to say that now you know how to find out what one
metre is. The ruler’s length is not a description of some spatial magnitude. It sets
a standard to isolate some spatial magnitude. You cannot ask for a criterion to
determine the correctness of this standard, just as it makes no sense to ask how
you know that the standard metre is one metre long.
Importantly, this does not imply that these standards cannot be inaccurate.
The triangle diagram can be badly drawn; the portrait can be imprecise.
Depending on the context of evaluation, the referent can match no target and
so on. However, the question of accuracy conditions of some representational
standards should be strictly distinguished from the question of how we know
what these standards represent.
Let us apply these considerations to Wittgenstein’s example of a picture of
an old man walking up a steep path that looks as if the old man is walking
down. Two questions can be raised here. We can meaningfully ask whether the
 Thinking with Images 163

image accurately identifies the target. In the same way, we can meaningfully ask
whether the indication of the ruler represents the distance or the velocity in
some frame of reference in relativistic physics.
However, it is meaningless to ask how we know that the iconic content is the
man walking up, not the man walking down. The iconic content permits both
interpretations, for it exemplifies both construction rules. These rules determine
the standards of interpretation. There can be no question of how I know which
rule it exemplifies, for we directly see the rule in the picture.
Think about how we interpret the content of the Necker cube. Depending
on the applied construction rule, it can be interpreted differently. However,
this does not imply that its content is indeterminable but only that it can be
interpreted differently. Moreover, it is meaningless to ask how we know that
we see one interpretation of the Necker cube and not another if the shape of
the Necker cube remains the same. The only answer is that we see it as such.
The construction rule we identify in picture-perception determines the way we
interpret the picture.
By the same token, Wittgenstein’s picture of an old man can be interpreted
differently, depending on the identified construction rules. However, there can
be no question about the criterion we use to identify the construction rule since
construction rules determine the interpretation standards.
Iconic content is directly referential. We cannot ask how we know that we
recognize an old man as walking up or down. We see it as such. By identifying
the construction rules, we directly see something in the picture. This is the
reason why it may appear that iconic content is primitive and unanalysable.
Thus, in the 2-D model, Wittgenstein’s argument appears to miss the mark.
Iconic content is indeterminable in the sense that it is criterion-less and
directly referential. This does not imply that iconic content is more primitive
or basic than propositional content. The rules of construction must be
identified in the picture to recognize an object. Such identification requires
skill and practice.
Moreover, holding that iconic content determines the standard of object
identification does not imply that the standard cannot be wrong. It can be
inaccurate, depending on what and how it is being represented.
The criterion-less and direct character of iconic content means that the
latter sets up the criterion for recognizing objects, properties and events. Iconic
content does not require any criterion of interpretation, for it determines this
criterion. Images represent how the world can be perceived in the same way as
rulers show how objects can look in some measurement set-up.
164 Thinking in Images

Metaphysical constraints

The Metaphysical Challenge concerns the problem of systematicity and

compositionality of thoughts. According to the challenge, images are not
systematic and lack canonical decomposition. Therefore, they cannot be bearers
of thoughts.
The backbone of the Metaphysical Challenge is the belief that images have
no logical structure, for the division of images into representational parts is
arbitrary (Casati and Varzi, 1999).2 Let us recall Fodor’s Picture Principle. It holds
that if P is a picture of X, then parts of P are pictures of parts of X. According
to Fodor (2008), all parts of a picture are its constituents. Take a picture of a
scene and cut it into arbitrarily chosen parts. Every picture part will represent a
part of the scene. Consequently, iconic representations do not have a canonical
decomposition, that is, a correct way to determine syntax and lexical primitives.
Fodor’s argument is based on the premise that every image-part is
representational. However, this premise is clearly false. Canvas texture is part of
a picture, but it does not represent the scene’s texture. The distances between
points are parts of a topological map, yet they do not represent the distances of
the mapped scene.3 What seems to be the problem?
First, Fodor operates with an ontologically restricted understanding of
the part concept. He takes it as referring to some events, material bodies as
well as spatiotemporal and geometrical regions. However, the concept of
part is ontologically neutral. It can refer to abstract entities such as properties,
propositions, types and kinds, too. For instance, one can hold that extraversion
is part of personality. Fodor’s premise that iconic representations can be divided
arbitrarily appears untenable with the ontologically neutral understanding of
the part concept. Take a gustatory image. It can be divided into property classes
such as salty, sweet, strong, minty but not in any way we choose.
Second, the Picture Principle premise appears acceptable only if we adopt an
unsophisticated version of the Traditional View, according to which pictorial
parts are pixel-like primitives. However, pixels are primitive when it comes to
information registration but not when it comes to representational content
(Burge, 2018). Analogously, phonemes are not representationally primitive from
the perspective of the sentence meaning; yet phonemes have to be registered to
form a sentence.
Luckily, the Traditional View is not the only game in town. According to
the 2-D model, iconic content consists of the construction rules by means of
which we identify the referent. Constructions have been defined as operations
 Thinking with Images 165

determining the parameters of some logical or physical space. The idea is that
we can distinguish between representational and nonrepresentational image
parts by identifying construction parameters.
Let us consider the triangle diagram. It represents the way a triangle can be
constructed. The triangle construction highlights the properties such as the
number of lines and the angle measure. These are the properties highlighted by
the construction parameters and content constituents. In contrast, properties
such as the colour of the lines are not representational as they are not singled out
by the construction parameters.
Image parts can be representational or nonrepresentational depending on the
identified construction rules. If I want to construct a congruent triangle, the size
of the lines is representational. If I want to construct a similar triangle, the shape
of the triangle is what matters. The size of the lines is nonrepresentational.
Let us compare this with the black hole picture. The construction rules
identify the crescent shape of the black hole emission ring. The properties of
the shape are representational because they are highlighted by the construction
parameters. In contrast, the photograph’s texture is nonrepresentational, for
there is no explicit construction rule in which the texture is highlighted by any
of its parameters.
Let me underline this point. In the 2-D model, there are no metaphysically
or psychologically predetermined representational primitives, such as lines
and dots (e.g. Burge, 2010, 2018; Camp, 2007; Tversky, 2004). Representational
primitives are the properties highlighted by the construction parameters by
means of which we identify a referent.
Depending on the identified construction rule, we isolate different
representational primitives. They are relative and context-sensitive. Consider
a picture of a red square. If one takes it as exemplifying red, the shape is
nonrepresentational. The colour is irrelevant if one takes it as exemplifying the
square construction.
Being relative and context-sensitive does not imply being arbitrary. The
meaning of indexicals is relative and context-sensitive, yet it is not arbitrary.
With this in mind, we can hold that images are canonically decomposable.
If images are canonically decomposable, then there is no reason to deny that
they are systematic. They clearly are. If I have an image of two objects o1 and
o2 where o1 is left of o2, I can easily imagine that o1 is right of o2. If I have an
auditory image, I can imagine the same music track with different paces and
pitches. Understanding one triangle diagram implies understanding any other.
Moreover, systematicity is a key concept in understanding how images work. Let
166 Thinking in Images

us recall the concept of natural generativity. One of the main tasks of any theory
of depiction is to explain how understanding one member of an iconic system
involves understanding any other member of the system. The concept of natural
generativity mirrors the concept of systematicity.
However, the systematicity of images cannot be explained in the same way as
the systematicity of propositional representations. Why is that so?
According to the classical explanation (e.g. Fodor, 1987; Fodor and Pylyshyn,
1988), propositional representations are systematic, for they have a logical form.
Propositional representations are decomposable into discrete atomic formulas
that contribute to the truth-conditions of the whole representation. For instance,
the meaning of the sentence ‘John is bald’ can be derived from the meaning of
its constituents. The idea is that every part is an atomic unit of a truth-evaluable
representation.
Moreover, propositional representations have combinatorial syntax, which
determines the place of their parts in this structure and the logical combinations
of content constituents. For instance, the sentence ‘John loves Mary’ consists
of the arguments ‘John’ and ‘Mary’, and the relation R of loving. Exhibiting the
thought that John loves Mary involves understanding R, which, in turn, makes it
possible to exhibit the thought that Mary loves John.
In contrast, images are deprived of logical form. They cannot be decomposed
into atomic units of truth-evaluable representations and lack combinatorial
syntax. Thus, we need an alternative explanation of the systematicity of images.
In the 2-D model, understanding iconic content involves knowing the
construction rules by means of which we identify a referent. Recognition-based
identification has been defined in terms of localizing the construction invariants
connected with knowing the construction rules that determine permissible
transformations of the constructed object. The idea is that we can think of the
systematicity of images in terms of preserving construction invariants. Two
representations are systematically co-related with each other if they preserve the
same construction invariants. Let me illustrate this with two cases.
Think about a triangle diagram. If I understand how to construct one kind
of a triangle diagram, I understand how to construct any other kind since I
understand a rule which determines what properties have to be preserved in those
constructions. Moreover, suppose I hold that I know how to construct a right
triangle diagram, but I do not recognize the construction of an obtuse triangle
diagram. In that case, it indicates that I do not understand triangle construction.
Let us consider an image of two objects o1 and o2 being left and right,
respectively. Understanding that o1 is left of o2 implies that I understand that o1
 Thinking with Images 167

can be right of o2, since they preserve their locations when the reference point is
switched. Additionally, if I hold that I understand the image where o1 and o2 are
left and right, respectively, but I cannot recognize the content of an image with
the switched reference point, then I do not understand the first image.
Moreover, by manipulating construction parameters, one can indicate
preserved properties. For instance, in the mental rotation tasks, we can spatially
transform a figure, then check whether its shape fits another. By the same token,
we can change a music track’s pace and pitch, preserving the sound sequence.
The shape of the spatial figure and the sequence of sounds are the invariants of
these constructions.
The manipulation of the construction parameters can be cognitively
productive. Think about a freehand drawing. It is based on searching for the best
measure to identify a referent. By determining the parameters of the drawing,
we identify different construction invariants, which, in turn, localize different
properties of the desired object.
What is more, the result of such manipulation cannot be deduced. According
to the measurement-theoretic account, images like rulers and balances identify
the results of the measurement procedures. By systematically manipulating
construction parameters, we test the consequences of some background
assumptions involved in image production and interpretation. These
consequences cannot be inferred. If they could be, any measurement would be
reducible to inferential procedures.
Thinking with images is, using Dennett’s metaphor (2013), ‘turning the knobs’.
It is about localizing objects, properties and events within some measurement
set-up. It involves finding out what would change if the construction parameters
were different. For instance, what would change if we spatially rotated an object in
the mental rotation tasks or how would the world look if we reversed its colours.
To sum up, images are systematic and decomposable. They are decomposable
into parts highlighted by construction parameters. They are systematic by means
of preserving construction invariants. We do not need to introduce propositional
structures to explain the systematicity and compositionality of images. Images
are systematic and decomposable in their own right.

Application 1: Representational format

In the previous sections, I explained how the 2-D model addresses the
epistemological, semantical and metaphysical challenges. In this and the
168 Thinking in Images

next section, I describe the metaphysical landscape that can be seen from the
measurement-theoretic account of images. I discuss two problems: the problem
of the mental representation format and the nature of mental imagery.
A mental representation format is the way information is organized in mind.
A discussion of mental representation formats addresses how the information in
our mind is stored and processed. It concerns the structure of representations
interpreted as a set of representational primitives and combinatorial principles.
Thus, to describe a representational format, one has to describe the structure of
representation: the primitive elements and the possible operations that can be
carried out with them.
Following the influential tradition, we distinguish between analog and
digital formats of representation. According to Goodman (1976), to be an
analog representational system means to be a dense representational system.
An example of a dense representational system is an old-fashioned clock
that represents time continuously, unlike a digital clock that represents time
discretely. Moreover, an analog representational system is relatively replete.
A representational system is relatively replete if, in comparison with other
systems, many features of its members are relevant to determining what they
represent. The system of old-fashioned analog clocks is not replete since only
the position of the clock’s hands matters. In comparison, in the case of images,
such features as colour, shape and size are relevant. However, for reasons that
will not be covered here (e.g. Kulvicki, 2006; Maley, 2011), it is doubtful whether
Goodman succeeded in adequately explaining analog and digital formats of
representation.
In the last fifty years, the distinction has been variously interpreted and
explicated (e.g. Fodor and Pylyshyn, 1981; Haugeland, 1998; Lewis, 1971;
McGinn, 1989; Peacocke, 2019). Across those approaches, digital representations
are generally understood to be discrete entities. Numerals provide a good
example. 0 and 1 are discrete because they indicate distinct and separable entities.
For every representational token, it is clear which type it instantiates. In contrast,
analog representations do not admit definite type-identity. For example, the
colour value of a given colour patch is measured on a continuous rather than a
discrete scale (Dretske, 1981). Iconic representations are believed to be analog
structures, which means that they are indiscrete. The structure of propositional
representations is believed to be based on discrete structures.
An alternative way to interpret the analog–digital distinction can be framed
in terms of constraints that the representational system puts on representational
content. Different representational systems put constraints on the content a
 Thinking with Images 169

representational system can carry and the range of possible transitions between
different contents. So, for example, an analog representation can represent a
magnitude value but not an integer (Beck, 2015) and so on.
If one wants to explain why some representational systems put constraints on
representational content, one has to describe the features of the representational
structure. For instance, one can explain why the Arabic numeral system is
preferred over the Roman numeral system by pointing out the fact that the Roman
numeral system does not include a zero. Analogously, iconic and propositional
representational systems put constraints on their representational content. A
theory of representational format should explain where these constraints come
from.
More generally, the question is what the concept of the mental representation
format should explain and how it can do that. Here, I claim that the problem
with the analog–digital distinction is not that it is not clear. Even if it were clear,
it would still be doubtful whether it could explain what it should explain, namely,
the difference between iconic and propositional representations.
There are at least two ways of describing the difference between propositional
and iconic representations. The first way refers to the idea of the informative
richness of images (e.g. Kitcher and Varzi, 2000). According to this criterion, iconic
representations display more information than propositional representations,
making images more effective in some reasonings (e.g. Larkin and Simon, 1987).
The analog–digital distinction seems to explain this difference, for the analog
representations consist of fuzzy sets of information, making iconic content vague
and rich. In contrast, digital representations consist of classical sets of information,
which makes propositional content more precise but informatively poorer.
However, a more profound understanding of the difference between
iconic and propositional content can be described in terms of their relation
to truth. Images are not truth-evaluable and lack logical form. In contrast,
propositions can be true or false and express logical relations. A full-fledged
theory of the mental representation format should explain what properties of
the representational structure are responsible for these constraints put on the
content.
Does the analog–digital distinction help us understand the difference between
iconic and propositional representations characterized in terms of their relation
to truth? It seems that it does not, as the iconic-propositional distinction does
not overlap with the analog–digital one (e.g. Peacocke, 2019).
If we characterize content in terms of its relation to truth, then the distinction
between analog and digital format is irrelevant for determining whether we are
170 Thinking in Images

dealing with a representational system that is iconic or propositional. As von

Neumann (1958) demonstrates, propositional operations, such as computations,
can be functions of digital and analog processes. Propositional representations
can be encoded in either digital or analog format. And although it does not
show that the distinction between analog and digital representational systems
is useless, it demonstrates that this distinction is insufficient for distinguishing
images and propositions.
The measurement-theoretic account offers an alternative description of the
representational format. In the 2-D model, images are measurement devices
used to localize the depicted object. This feature generates some consequences
for our thinking about the representational format. In the measurement-
theoretic perspective, at least two marks distinguish the structure of iconic
and propositional representations. The first concerns the problem of context
sensitivity. The second is the problem of predication.
First, mechanisms of information processing in iconic representations are
domain-specific. They are domain-specific if the operations defined for the
structure elements depend on the application area. If the operations do not
depend on the area they are applied to, the relevant mechanisms are domain-
general. Such is the case of the rules of addition; regardless of what one is adding,
the rules are the same. In contrast, measurement devices are domain-specific as
measurement methods depend on what and how you measure.
Similarly, processing iconic representations depends on the modality and
vehicle of representation.4 For instance, visual and gustatory representations
of wine belong to two different systems of representation. Moreover, there is a
significant difference between imagining tasting wine and actually tasting it. The
information is processed differently, depending on what and how it is represented.
In contrast, propositional representations are domain-general. The
way information is processed is insensitive to the modality and vehicle of
representation. For instance, the proposition wine is red remains the same
regardless of whether it is expressed in speech or writing, English or German.
Second, unlike propositions, images are not predicative. When I hold that
snow is white, I attribute the property white to the object snow, where the terms
‘snow’ and ‘white’ work as arguments of a predicative function expressed in the
proposition.
The predicative nature of representation distinguishes the proposition from
an image. The proposition carries the denotations of the terms into a truth-value
(Rescorla, 2009b), while the image of the white snow does not. What properties
of the representational format can explain this?
 Thinking with Images 171

For one thing, predication is based on a compositional and combinatorial

mechanism. The compositional and combinatorial character of a propositional
system means that if I have the propositions snow is white and a triangle
is red, I can form the structurally similar propositions snow is red and a
triangle is white. Moreover, if I know that the proposition snow is white
is true, I can infer that the proposition snow has a colour is also true. This
means that an output of one operation of predication can be an input of another
higher-order operation. These operations are hierarchically organized (Camp,
2018). From the proposition snow is white, I can infer that snow has a colour,
but from the proposition snow has a colour I cannot infer that snow is white.
For another, the output and input information comes in discrete chunks.
Consequently, the proposition snow is white attributes the property whiteness
and no other property to snow. It says nothing about the hue of the colour or the
shape of snow, for there is a one-to-one correspondence between a vehicle and
content. Every chunk of information needs a separate vehicle to be expressed.
Thus, propositional representations are hierarchically organized and based on a
structure made of discrete chunks of information.
In contrast, iconic representation is neither hierarchically organized nor based
on discrete chunks of information. Hierarchical organization of representational
structure means, for example, that the proposition snow is white implies snow
has a colour but not the other way round. In the case of iconic representation,
both pieces of information are processed simultaneously. An image of white
snow represents both that snow is white and that snow has a colour.
The non-hierarchical organization of information is often confused with the
holistic nature of the components of the iconic structure. These two concepts,
however, have to be separated since we can have non-hierarchical information
architecture based on non-holistic components. For instance, parallel computing
is non-hierarchically organized and based on discrete chunks of information.
The holistic nature of representation indicates the relation between the
content and the vehicle of representation. To my knowledge, the first person to
draw attention to this idea was Twardowski (1965), who ascribed the feature of
concreteness to imaginings. He understood concreteness as the combination of
multiple properties in a single representation. Similarly, the concept of holism
is understood (e.g. Green and Quilty-Dunn, 2021; Kulvicki, 2020) as the thesis
that multiple pieces of information expressing the content of a representation
are assigned to the same vehicle of representation.
This means that there is no separate vehicle for every chunk of information
corresponding to different representational properties. In other words, there is
172 Thinking in Images

no one-to-one correspondence between parts of the information and parts of the

vehicle. For instance, the part of an icon that represents the colour of a triangle is
the same part that represents its shape.
In the 2-D model, the non-hierarchical information architecture and holistic
nature of representational primitives result from how images represent the world.
Unlike propositions, images represent by exemplifying the rules of construction
which identify the represented object. In short, images are a kind of sample of
the represented object.
Being a sample suggests that images have to be able to exemplify many
properties at once. Depending on the construction rule we pick out, different
representational properties are highlighted. Yet, there is no principled way to
determine the property that is represented by the image. That requires that the
vehicle carry more than one piece of information at once.
If there is no one-to-one correspondence between the vehicle and
information, then information processing cannot be hierarchically organized.
For instance, a triangle diagram represents the triangle as being right and red,
simultaneously. Such information is not inferred. It is displayed by the same
vehicle of information.
Thus, the format of iconic representations is domain-specific; it processes
information in a non-hierarchical manner and is based on holistic components.
The structure of propositional representations is domain-general; it processes
information hierarchically and is based on discrete chunks of information.
Does this quality of representational format allow us to explain the constraints
put on the content of images and propositions? It does. First, displaying multiple
pieces of information by one vehicle of representation explains why we hold that
images are informatively rich. Iconic content is informatively rich, for it is able to
express more information than propositions, which are constrained by the one-
to-one correspondence between the vehicle and the information.
Second, being truth-evaluable and having a logical form requires
propositional structure. It involves a special kind of relation between the vehicle
of representation and the content. To speak of truth, we need to be able to
distinguish atomic formulas that can satisfy T-schema. That requires a one-to-
one correspondence between the vehicle and the information. Similarly, to infer
from a is F and if a is F, then a is G that a is G, one has to be able to assign
distinct content to the vehicles of representation expressed by logical variables.
Moreover, the chunks of information have to be hierarchically organized. From
the thought that a is G and that if a is F, then a is G I cannot infer that a is F.
Propositional representations are hierarchically organized.
 Thinking with Images 173

In contrast, images are organized non-hierarchically. The information is

processed simultaneously. When I see a picture of white snow, I see that it has
a colour; when I see a colourful picture, I can see the specific colour of the
picture.
Moreover, propositions require a representational structure that can abstract
away from the nature of the vehicle of representation. For instance, the reasoning
if A then B and A then B is correct regardless of whether it is conducted mentally
or on paper. Propositional representations meet this requirement since they are
domain-general. In contrast, iconic representations are domain-specific. The
nature of the iconic representation vehicle affects how information is processed.
For instance, tasting wine and imagining tasting wine are informationally two
different representations.
To sum up, iconic representations are not truth-evaluable and lack logical
form. These facts are easier to understand if we hold that iconic representations
are domain-specific, that they process information in a non-hierarchical
fashion and that their structure is based on holistic components. In contrast,
propositional representations are domain-general, they process information
hierarchically and their structure is based on discrete elements.

Application 2: Mental imagery

Mental imagery is usually understood as a faculty of bringing about mental

images and seeing with the mind’s eye (e.g. Kosslyn et al., 2006; MacKisack et al.,
2016).5 However, asking about the nature of mental imagery is tricky.
In cognitive psychology, the nature of mental imagery has been the
subject of the so-called imagery debate. However, there has never been one
understanding of the debate’s subject. Instead, there have been at least three
parallel debates. What started as a discussion on epistemological functions
of mental images (Paivio, 1963; Shepard and Metzler, 1971) soon turned –
mainly thanks to the computational character of research in early cognitive
science – into a discussion on the format of mental representation. On the
one hand, pictorialists (Kosslyn, 1980) have held that the format of mental
images is perceptual-like. On the other, descriptionalists (Pylyshyn, 1973)
have argued that mental images are formed out of structured descriptions.
Most descriptionalists have argued that mental images are epiphenomena of
some internal language-like processes. From the beginning of the 1990s, the
discussion has focused mostly on neurobiological mechanisms responsible for
174 Thinking in Images

generating mental images. None of these debates has direct implications for
others.
Nowadays, the imagery debate is mostly considered dead (Pearson and
Kosslyn, 2015). In cognitive psychology, there is a consensus that descriptionalists
fail to explain the perceptual nature of mental images. Moreover, it is held that
mental imagery is a form of weak and off-line perception (e.g. Pearson et al.,
2015). According to the weak perception theory of mental imagery, mental
images are a form of less vivid perception not triggered by sensory input.
The claim that descriptionalism is a dead end seems to be noncontroversial.
Moreover, there is no doubt that imagining is a form of perceptual experience
and that the neural correlates of perception and mental imagery are common
to a significant degree (e.g. Kosslyn et al., 2006; Nanay, 2015; Pearson et al.,
2015).6 There are, however, doubts regarding the consequences of these claims.
Primarily, the weak perception theory of mental imagery is philosophically
dubious.
First, although mental imagery and perception share the mechanisms
responsible for generating representations, there is an asymmetry between the
content of mental imagery and perception. It is believed that the content of
mental images is displayed as already interpreted (e.g. Chambers and Reisberg,
1985; Hinton, 1979; Ishai et al., 2000; Ittelson, 1996; McGinn, 2006; Reisberg,
1996; Reisberg and Heuer, 2005; Sartre, 1962; Scaife and Rogers, 1996; Slezak,
1995). In contrast, we can always reinterpret the content of perception.
The most striking illustration of this fact is the problem with reinterpreting
the so-called bistable figures (such as the duck-rabbit picture) in mental imagery.
Although reinterpretation of mental images is possible (e.g. Reisberg and
Chambers, 1991; Finke et al., 1989; Peterson et al., 1992), it is more cognitively
loaded and significantly less efficient than in perception (Mast and Kosslyn, 2002;
Kamermans et al., 2019). For instance, in the studies of Reisberg and Chambers
(1991), subjects were told to remember and imagine a figure with the same
geometry as the state of Texas rotated 90° clockwise. Even when they imagined
rotating the figure mentally so that it assumed its canonical orientation, they
were still unable to discover that it represented Texas. In contrast, the task was
trivial when subjects were told to rotate the image being seen.
The asymmetry between imagery and perception when interpreting bistable
figures cannot be explained by holding that mental images are displayed in the
mind’s eye too briefly, which might suggest that imagery is affected by memory
limitations (see Kosslyn, 1994; Kosslyn et al., 2006). For one thing, reinterpreting
the perceptual content does not have to be time-consuming. Often, we can
 Thinking with Images 175

switch between two interpretations instantaneously. For another, the time that
affects the efficiency of reinterpretation tasks is the same time that suffices to
solve mental rotation and mental zooming tasks in Kosslyn’s classic research
on mental imagery (e.g. Kosslyn et al., 1978). Therefore, the time for which
information is available does not seem to be a relevant factor.
Second, suppose that we respond to the first doubt by holding that mental
imagery is less vivid than perception. However, that is a position that we do
not want to keep if we want to defend the imagistic theory of thinking. For
one thing, the weak perception theory of mental imagery appears to blur the
distinction between thoughts and perception and is vulnerable to arguments
from cognitive impenetrability of perception (Cavedon-Taylor, 2021; Firestone
and Scholl, 2016; Raftopoulos, 2009). For another, it seems to take us back to the
neo-Lockean positions we previously rejected (see Slezak, 2002a). In fact, the
weak perception theory of mental imagery is no more philosophically attractive
than the Lockean theory of ideas, which has been a philosophical dead end
for over three hundred years. If we hold that mental imagery entails weak
perception, then either exercising mental imagery is not a thinking operation
or speaking of imagistic theory of thought is senseless. Both alternatives are
unattractive.
Where do we stand? The situation resembles the one described by Goodman
and Elgin (1988) as the dilemma of mental imagery. Either we accept the weak
perception theory of mental imagery and leave out the questions asked by
philosophers and other troublemakers, or we just dismiss the talk of mental
images, holding that they are just the epiphenomena of some more substantive
cognitive mechanism. I think that both horns of the dilemma should be rejected.
The problem we face stems from overusing the photographic metaphor in
thinking about mental imagery. According to this metaphor, mental imagery
is like a camera, and mental images are like photographic snapshots displayed
before the mind’s eye. The force of the photographic metaphor results from
our attachment to the Traditional View. If images are copies of the world, then
mental images are copies of perceptual experiences. That is, however, where we
get stuck in conceptual confusion. There is no Cartesian cinema in the head;
some inner eye does not see mental images.
According to Block (1983b), the photographic model of mental imagery
should be replaced by the one in which mental images are based on mechanisms
involving constructive processes. Mental images are more like drawings rather
than photographs. However, Block has never been specific enough to explain
what such processes can be.
176 Thinking in Images

In the measurement-theoretic account, the drawing model of mental

imagery can be explicated as follows. According to the 2-D model, images are
measurement devices. They exemplify the construction rules by means of which
we identify the depicted objects and events. Iconic content localizes objects and
events in some physical or logical space by determining the parameters of the
space. Knowing the iconic content is knowing the construction rules that set
these parameters.
The general idea is that we can distinguish between two ways of understanding
the concept of the mental image. One refers to having a conscious, perceptual-
like experience of an object.7 In the 2-D model, it is a vehicle of representation
that exemplifies iconic content. Let us call this the experiential account of
mental images. The other refers to the iconic content that is exemplified by this
experience. Let us call this the constructive account of mental images. Having a
mental image in the constructive sense is knowing the iconic content. According
to the measurement-theoretic account of images, this means knowing the
construction rules through which we identify the object. Just like having a
concept does not imply producing it, having a mental image in the constructive
sense does not imply producing the experience of mental images.
Constructive understanding of the nature of mental images has at least three
metaphysical consequences for our understanding of what mental imagery is.
First, mental imagery as a faculty of bringing about mental images is not, as it
were, a moving theatre where mental images are displayed before the mind’s
eye. The cinema-in-the-head idea of mental imagery results from taking the
experiential understanding of mental images as constitutive of what mental
images are. Mental images, however, do not have to be experienced. Mental
images as conscious experiences are exemplifications of a more basic iconic
mechanism.
According to the constructive understanding of mental images, having
a mental image is an ability to perform certain activities (Goodman, 1984;
Goodman and Elgin, 1988). In the 2-D model, to have a mental image is to
know a construction rule that is a measure by means of which we identify
objects and events. For knowing this measure, however, it is irrelevant
whether you have an object that exemplifies it. It is crucial to have the skill of
measuring. For instance, you do not need a ruler to measure distance. If you
know how to measure distance, you can use anything that can help you, for
example, a stick.
Mental imagery is the skill of applying construction rules to identify objects
and events in some space. It involves determining the parameters of some space
 Thinking with Images 177

to arrive at the constructed object and recognize it. Mental images as conscious
experiences are exemplifications of these skills.
Does this mean mental images as conscious experiences are irrelevant to
mental imagery? It depends on how you understand this question. Thinking
of mental imagery in terms of the skills of producing mental images does not
imply that it actually creates them. By the same token, when thinking in words,
you do not have to produce these words. For instance, you can have a thought
and search for the words to express it. What is crucial is the skill to identify the
linguistic representations that correctly express this thought.
Mental images as conscious experiences are products of mental imagery taken
as a procedural knowledge of construction rules. This knowledge manifests itself in
the operations of constructing and interpreting the mental image. However, it can
also be manifested by any intelligent use of pictures, such as skilful use of diagrams.
Taking (mental) images as products of procedural knowledge does not imply
that they are irrelevant for exercising such a skill. If one cannot identify the
construction rules represented by an image, regardless of whether the image is
mental or extramental, then one does not have this procedural knowledge. In
this sense (mental) images are necessary for mental imagery.
To illustrate this, let us once again consider mental rotation tasks. The idea
is that subjects manipulate the parameters of perceptual space by rotating the
figure in order to localize construction invariants. Finding these invariants solves
the task of identifying two similar objects. The task cannot be solved without
mental imagery since it involves procedural knowledge of how to determine
these parameters, for example, setting the direction of rotation.
However, having such procedural knowledge is dissociated from having an
actual experience that exemplifies it. The operation of rotating a figure can be
visualized in the form of a conscious experience, but it does not have to be. We
do not have to represent the skill as a conscious mental image to exercise it. We
have to be able, however, to recognize the identity of figures when we visualize
them or display them in pictures.
Second, if mental imagery does not have to be conscious, then we can speak
meaningfully about unconscious mental images. Let us recall the phenomenon
of aphantasia, which is the lack of the ability to form mental images (e.g.
Keogh and Pearson, 2018; Milton et al., 2020; Zeman et al., 2015, 2020). This
phenomenon appears to be theoretically problematic only to the experiential
accounts of mental imagery. From the measurement-theoretic perspective, the
phenomenon of aphantasia does not seem to be particularly mysterious. What is
lacking is not mental imagery but the awareness of having mental images.
178 Thinking in Images

Being unaware of a mental image does not imply that one lacks the skills
to apply construction rules. For example, aphantastics can recognize identical
figures represented in pictures which indicates that they do not lack mental
imagery.
Third, the relationship between perception and mental imagery seems much
more profound than the weak perception theory suggests. Mental imagery is
not some less vivid perception. It is a skill of perceiving based on procedural
knowledge of construction rules. It involves the skill of identifying the parameters
of perceptual space in order to localize an object or event.
Thinking of mental imagery in terms of perceptual skills allows us to explicate
the relation between perception and mental imagery. For one thing, it allows us
to acknowledge the perceptual nature of mental images. If mental imagery is a
skilful use of perception, then the neural mechanisms of perception and mental
imagery should be shared, which is indeed the case. Moreover, if the format of
the iconic representation is domain-specific, then we should not expect to have
one mechanism responsible for mental imagery, which is also the case. There is
no common mechanism of mental imagery for all modalities.
For another, we can explain the asymmetry in interpreting bistable figures in
imagery and perception. If mental images are representations of measures, then we
expect their meanings to be fixed. They are not open to interpretation, in the same
way as the meaning of a standard metre is not open to interpretation. Images set
the conditions of interpretation by exemplifying the construction rules by means
of which we identify objects. In contrast, perceptions are open to interpretation,
for it is always possible to ask which measure should be applied to interpret them.
Depending on different measures, we highlight different perceptual properties.
Moreover, interpreting mental imagery in terms of a set of skills gives us
a better picture of the relation between perception and thinking than the neo-
Lockean approaches. On the one hand, it offers a hint on how to think about the
role of mental imagery in perception. A good illustration of such a role is the
so-called amodal completion. It is the common phenomenon of perceiving the
parts of perceived objects from which we have no sensory stimulation (e.g. Nanay,
2018).8 For example, when seeing a cat behind a fence, we amodally complete the
parts of the cat that are occluded. Importantly, amodal completion is a perceptual
phenomenon. It happens very early in the sensory cortices (Thielen et al., 2019).
In a sense, we see the occluded parts that are not triggered by the retinal input.
Now, if we think about mental imagery in terms of perceptual skills, then
amodal completion is a form of mental imagery. The procedural knowledge of
 Thinking with Images 179

how to construct an object amodally completes this object in perception. For

instance, seeing a part of a cat hidden behind a fence involves applying the
construction rule by which we reconstruct the whole object in mind. By the
same token, when seeing an incomplete or badly drawn triangle, we complete
the figure, for we know how to construct triangles.
On the other hand, the measurement-theoretic account appears to be
invulnerable to objections from the thought-perception border. According
to this objection, there is no room for any representation to mediate between
thoughts and perception since they possess mutually incompatible properties.
However, mental imagery is not a representation that mediates between thought
and perception. It is a perceptual skill of applying construction rules that results
in recognition-based identification of objects.
Recognition-based identification triggered by knowing specific construction
rules is a phenomenon that is both perceptual and cognitive (Abid, 2021). It
is perceptual since it involves perceiving the recognized objects. It is cognitive
since it involves a cognitive operation of applying the rules of construction. And
yet, it is not a belief-based operation. It does not involve having any beliefs about
the recognized objects. Having beliefs involves possessing concepts and the
language to express them. Perceptual recognition is a skill that precedes having
beliefs.
Let us go back to amodal completion. It is both perceptual and cognitive.
From the neuroscientific point of view, we see unperceived objects. However, it
is not the operation that has to involve any beliefs. I can amodally complete a
perceptual object I have no beliefs about. Instead, it is a matter of a thoughtful
exercise of skill of recognition.
To sum up, by distinguishing between the experiential and constructive
understanding of mental images and interpreting mental imagery in terms of
constructive skills, we can solve the dilemma of mental imagery. On the one
hand, mental imagery as a Cartesian theatre displayed before the mind’s eye is
only another instantiation of the myth of Ryle’s (1949) ghost in the machine.
On the other, mental imagery is a skill of applying construction rules to identify
objects in some space. In this second understanding of the term, mental
imagery is an empirically well-grounded phenomenon that can be studied by
mental rotation, mental zooming tasks and so on. From the philosophical point
of view, mental imagery seems to be no more problematic and can be studied
with empirical methods. If some philosophical questions can be answered with
empirical methods, the philosophical job is done.
180 Thinking in Images

Summary

In this chapter, I explain how the measurement-theoretic account of images

gives us an insight into the nature of imagistic knowledge. Images exemplify
measures that make it possible to understand the objects of our beliefs and
theories by means of localizing these objects in some logical or physical space. I
hold that iconic content does not require criteria of interpretation. In contrast,
it sets these criteria. Images, like rulers and balances, exemplify the measures by
which we recognize objects, properties and events. Finally, I demonstrate that
images are systematic and decomposable in a different way than propositional
representations. They are systematic and decomposable as they exemplify the
rules of construction which identify construction parameters and invariants.
I show the measurement-theoretic account of images allows us to acknowledge
the irreducible role of images in thinking. Images serve the function of localizing
objects of our beliefs and theories. At the same time, imagistic and propositional
representations are bound together. Although images can localize objects,
enabling their recognition-based identification, they cannot describe them. We
need propositional structures to make images meaningful, that is, to place them
within our beliefs and theories.
Last but not least, I demonstrate the consequences of the measurement-
theoretic account of images for our thinking about the mind. First, the
account allows us to rethink the concept of the representational format.
Iconic representations are domain-specific, they process information in a
non-hierarchical fashion and their structure is based on holistic components.
Second, it offers us a fresh perspective on the idea of mental imagery. According
to the measurement-theoretic account, mental imagery is a measuring skill that
determines the parameters of perception.
 8

Conclusion

In this book, I have been trying to solve the philosophical problem of

determining the nature of thinking with images. The problem is how to reconcile
two separately legit but mutually incompatible theses. On the one hand, we are
justified in holding that images are irreplaceable in cognitive practice. On the
other, the way we think about thinking makes no room for imagistic thoughts.
Images seem to be instrumental and peripheral in thinking.
The problem stems from two sources. First, it is based on our assumptions
regarding thinking. We usually hold that thinking is propositional and action-
based. Thinking is an operation carried out on propositional structures that are
expressed in language-like representations. It builds the so-called Received View
on the nature of thinking.
Second, when thinking about images, we usually adopt a version of the
so-called Traditional View, according to which images represent objects by
means of resembling them. Adopting the Traditional View, however, makes
images vulnerable to Wittgenstein’s argument from the indeterminacy of content
and Frege-Davidson’s argument from lack of logical form. Thus, in contrast
to propositional representations, images seem unable to deliver the theory of
knowledge and the content of thought. They cannot explain the systematic and
decomposable structure of thought. Therefore, imagistic thinking seems to be a
contradiction in terms.
And yet, we cannot simply reject the Received and Traditional Views. They
both express intuitions that are central to our thinking about the nature of
thoughts and images.
Throughout the book, I have been trying to address these problems by taking
the so-called operational approach, according to which the question ‘What is
thinking with images?’ can be taken as the question of possible operations carried
out with the help of images. Explaining the nature of imagistic thinking means
182 Thinking in Images

pointing at the conditions that must be met if such operations are possible. Thus,
I investigate what images do and try to find something that does it.
To find out what images do, I analyse the functions of images. I am doing
so by presenting two case studies. The first one concerns the knot diagrams in
mathematics. With this example, I study the idea of the content of diagrams. To
explain the way diagrams represent, I introduce the concept of construction.
It refers to the procedures of arriving at a goal by determining the parameters
of some logical or physical space. The general idea is that diagrams represent
objects and events by representing the ways these objects and scenarios can be
constructed.
By introducing the concept of construction, I am able to offer an alternative
account of iconic content. In contrast to the Traditional View, I hold that
resemblance works as an explanandum but not as an explanans. I hold that the
concept of construction can explain why images resemble their objects, avoiding
the objections directed at resemblance-based semantics.
In the second case study, I analyse the black hole picture. To explain the way
this picture represents its object, I introduce the concept of recognition-based
identification. I argue that recognition-based identification is different from
demonstrative reference and descriptions. Recognition is based on identifying
the construction invariants.
Based on the concepts of construction and recognition-based identification,
I introduce the so-called two-dimensional model of iconic reference. According
to this model, images denote their targets and exemplify the rules of construction
by means of which they reidentify the referent.
I argue that the best way to explain these properties of iconic reference is to
think of images in terms of measurement devices. Just like rulers and balances,
images represent the ways of localizing objects and events by determining the
parameters of some space. In contrast to the Traditional View, images are not
copies of the world. They are the measurement devices we use to localize the
properties of the world.
Next, I demonstrate that the measurement-theoretic account of images
offers a theory of imagistic knowledge and content. In the 2-D model, imagistic
thoughts are systematic and decomposable. Thus, thinking of images in terms of
measurement devices meets the requirements of a theory of thought. Moreover,
it is a kind of thinking that is irreducible to propositional thinking.
Last but not least, I show the metaphysical consequences of the measurement-
theoretic account of images for our thinking about the format of mental
representations and mental imagery. I hold that we can replace the analog–
 Conclusion 183

digital distinction with one where the structure of iconic representations

is domain-specific, non-hierarchical and based on holistic components. In
contrast, propositional representations are domain-general, hierarchical and
based on discrete elements.
Finally, according to the measurement-theoretic account, we can abandon
the Cartesian view on mental imagery. In the 2-D model, mental imagery is a
skill of applying construction rules that determine the parameters of perception.
To sum up, I have been arguing that the imagistic theory of thought can be
made meaningful if we think of images in terms of measurement devices. In
the measurement-theoretic perspective, images can be seen as irreducible to
propositional representations. This does not imply, however, that images are
more basic than propositional representations or that they can ground the
meaning of propositions. Iconic and propositional representations have distinct
functions, yet they are entangled in thinking. Thus, if one asks whether we think
with words or images, the answer would be that we think with both. Thinking is
both iconic and propositional.
Does this imply that we have words or images in the mind? No, it does not.
Thinking is an operation that can be expressed with either words or images.
Words and images are necessary for thinking only in the sense that they are the
only medium that can express these operations. Images and words are necessary
for having thoughts, just as numerals are necessary for numbers. They represent
some operations. In the case of images, these are the operations of construction
and recognition. Propositions are indispensable for the operations involving
truth values, such as inference or having beliefs.
One of the consequences of the ideas presented here is that knowledge is more
than propositional knowledge and thinking is much more than the Fregean game
in truth. In contrast to Frege, I hold that thinking involves much more than truth-
evaluable operations. What this ‘something more’ is should be a philosophical
topic, just as truth is. Even if this book does not succeed in presenting a coherent
theory of non-propositional thinking, I hope that it presents the need for such a
theory. I am convinced that the philosophy of images can play the same role in
our understanding of the mind as the philosophy of language has played.
Although the book covers many topics, it does not cover all of them. I barely
touch on the problem of scientific models and the arts. I deliberately omit the
problems of metaphors and imagination. These issues require separate studies.
I believe, however, that they can be interpreted in line with the 2-D model and
incorporated into the imagistic theory of thought. Yet, any attempt to discuss
these problems in one book would far exceed the reader's patience.
184 Thinking in Images

In the book, I have argued that although the idea that we think with images is
intuitive, it has been rarely thought through. However, this does not mean that
it has never been considered. In general, no idea presented here cannot be found
in Kant, Peirce, Wittgenstein, Evans or Goodman. All I am presenting here is an
attempt to think through the consequences of their ideas. All I am hoping for
is that we can see some ideas clearer. That is the only thing a philosopher can
fairly offer.
 Notes

Introduction

1 I use the expressions ‘thinking with images’ and ‘thinking in images’ interchangeably.
2 See, e.g., Irvine (2014), Lakoff (1994), Lakoff and Johnson (1980), Langacker (1987).
For a contemporary defence of an imagery-based theory of semantics, see, e.g., Ellis
(1995), Lowe (1996), Thomas (1997) and Nyíri (2001).
3 For a philosophical and empirical defence of the existence of unconscious mental
images, see, e.g., Church (2008), Nanay (2010, 2015, 2017, 2021) and Zeman et al.
(2010, 2015). The discussion mostly concerns the phenomenon of aphantasia, which
is a condition where one does not possess the ability to visualize images (e.g. Keogh
and Pearson, 2018; Milton et al., 2020; Zeman et al., 2015, 2020) and form non-
visual mental images (e.g. Dawes et al., 2020).
4 Mental and extramental images undoubtedly differ in many respects. The
differences, however, should not obscure the similarities (e.g. Abell and Currie,
1999). Reasoning with the help of mental images is often performed interactively
with external representations (e.g. Kirsh and Maglio, 1994). Forming an extramental
image is not a post-hoc expression of some reasoning done with the help of
mental images but an integral part of this reasoning (Sheredos and Bechtel, 2020).
Moreover, an ability to form an external image assumes an ability to form a mental
image; for example, knowing how to construct a triangle on a piece of paper is
knowing how to construct a triangle in the mind, and vice versa (Maynard, 2011).
5 Most generally, this relation to a proposition can be described in Fregean terms as an
act of ‘grasping’ thoughts, which refers to every act of discovering or understanding
the meaning of a thought and which is equivalent to knowing the truth-conditions
of a proposition. Grasping a thought is distinguished from evaluating the truth value
of a thought, for knowing the conditions according to which a thought is true is not
equivalent to knowing whether it is true.

Chapter 1

1 In fact, it is almost impossible to find a single empirical article without any reference
to a graph or a diagram. According to the studies of Zacks et al. (2002), the number
of graphs in scientific journals doubled between 1984 and 1994. Nothing suggests
that this trend has not continued.
186 Notes

2 The concept of diagrammatic reasoning is vague, since there is no general agreement

on the definition of a diagram (e.g. Pietarinen, 2016; Shimojima, 2001; Stjernfelt,
2007). A diagram is usually interpreted as a spatial representation of abstract pieces
of information that enable us to infer the features of the represented objects through
inspection of the spatial characteristics of the representation (Swoyer, 1991).
Diagrammatic reasoning may be described in terms of a reasoning process based on
manipulation and inspection of diagrams.

Chapter 2

1 In cognitive psychology, the belief that thinking is propositional and action-based is best
expressed by Holyoak and Morrison’s (2012) definition of thinking as the systematic
manipulation of cognitive representations determining current or possible states of the
world. In accordance with the Received View, thinking is most often operationalized as
an act of reasoning and problem-solving (Evans, 2017). On the history of research on
thinking in cognitive psychology, see, e.g., Dominowski and Bourne (1994).
2 An example of the epistemological argument from the content indeterminacy of
images is Descartes’s well-known chiliagon-argument. According to Descartes, an
image of a chiliagon is indistinguishable from an image of polygon with 999 sides,
and yet we have clear and distinct ideas of a chiliagon and a 999-gon. Thus,
thinking cannot be imagistic. However, this argument is clearly invalid, since it
does not imply that our thoughts cannot be imagistic. It demonstrates only that
our perceptual capacities are limited. If we have enough time to draw a picture of
chiliagon and a 999-gon and count the angles, then it is possible to distinguish a
chiliagon and a 999-gon (see Dennett, 1990). In the imagery debate, contemporary
versions of Descartes’s epistemological argument (sharing the flaws of the original
argument) are Dennett’s argument (1981a) from a striped tiger and Armstrong’s
argument (1968) from a speckled hen.
3 Kulvicki’s argument concerns pictures, but it can be readily extended to all imagistic
genera.
4 The psychological instantiation of the pictorial fallacy is the so-called lack of
pictorial competence. It is an inability to discriminate between objects and
depictions of these objects which characterizes young children up to two years old.
It manifests itself in toddlers’ attempts to manually grasp the depicted objects as if
the real objects were presented (see DeLoache et al., 2003).
5 Gregory (2020) argues that images are non-propositional but predicative. The
Semantical Challenge shows that the predicative function is suspect.
6 According to Evans (1982), systematicity is constrained by semantic conditions
of appropriateness. For instance, thinking that John fell into the lake need
 Notes 187

not entail a capacity to think that the lake fell into John. However, even if a
well-formed string is semantically absurd, that does not mean that it cannot express
a thought. For one thing, we can entertain absurd thoughts. For another, absurd
but well-formed strings can serve as a basis of inferences in logic. See, e.g., Camp
(2004).
7 That does not mean that they lack construction rules. They are obviously rule-
governed. That is why if one understands how to interpret one Venn diagram, then
one understands how to understand another (e.g. Tversky, 2004). A representational
system can have construction rules without syntactic structure. A syntax requires
discrete symbols and logical rules that transform strings of symbols into well-
formed formulas of the system.
8 Propositional and associationist theories of thought, just as the Received View, do
not represent actual theories or a set of theories but certain ideal types or research
programmes. Although they share some methods and questions, they are sets of
related yet independent theses. In this general sense, associationist theories of
thought have to be distinguished from historical associationist theories of mind
such as Mill’s and Hume’s.
9 There are also intermediate strategies, such as Rescorla’s (2009).
10 This distinction does not cover all possible accounts of thinking. There are also
atypical positions, such as Ryle’s adverbial theory.
11 To the best of my knowledge, Kant and Peirce were the first to recognize perceptual
and spatial metaphors as complementary in their theories of thinking.

Chapter 3

1 In twentieth-century philosophy, some philosophers have seen no choice but

to directly include this intuition in their theories of mind. The most significant
examples were Russell (1921) and H. H. Price (1953). Both argued that a full-
blooded theory of mind should include imagistic thoughts and that images are more
basic than words. In a sense, images have a superiority over words, for they ground
the meaning of language. As a consequence, they, rather tacitly, take imagistic
thought to be a primitive and unanalysable phenomenon (e.g. List, 1981). In other
words, they assume that the content of images is self-explanatory. However, if
Wittgenstein’s argument from content indeterminacy is sound, then images cannot
be self-interpretable atoms of meaning.
We encounter similar problems in Arnheim (1969, 1980) and Wollheim (1984).
Both try to find a description of the phenomenology of thoughts that is in accord
with the imagistic nature of representations. However, even if some thought
processes may appear as if they were imagistic, that does not mean that their real
188 Notes

nature is imagistic. It only means that we need an explanation for why they appear
imagistic. Given that they demonstrate that we think with images, it does not bring
us any closer to answering the question of what thinking with images is.
In cognitive science, we encounter the same problem. It is well established
that there are individual differences in thinking and people can be characterized
in terms of being ‘concrete’ versus ‘abstract’ thinkers (e.g. Paivio, 1963; Paivio
and Harshman, 1983; Kozhevnikov et al., 2005). There is also something to the
observation that visual aids can facilitate the process of thinking. For instance, it
has been frequently indicated that the same sort of ‘visual’ style of thinking fosters
the creativity of scientists (e.g. Holyoak and Thagard, 1996; Johnson-Laird, 1998;
Otis, 2015; Rocke, 2010; Shepard, 1978). However, we must bear in mind that we
lack a clear-cut definition of thinking styles. This concept is most often understood
very broadly. It refers to many different domains, such as cognitive styles, learning
styles, personality types, and so on, which are not integrated within a general
theory of psychological functioning. There is strong evidence that there are no
unified visual and verbal cognitive styles (e.g. Kozhevnikov et al., 2005, 2010).
Moreover, it is crucial to remember that the psychology of individual differences
does not study thinking mechanisms but asks about individual differences in using
different mechanisms depending on preferences and needs. Thinking styles are
studied based on questionnaires and surveys. There is no way to study them using
neuroimaging methods. The fact that there are individual differences in solving
cognitive tasks does not imply that we were born with some inherent thinking
preferences. It can mean that using images is a good cognitive strategy for specific
tasks, and some individuals are more skilled in employing one strategy or another.
Thus, referring to thinking styles does not explain what thinking with images is. It
is only another way of describing the explanandum. A proper explanation requires
analysing the source of individual differences in cognitive tasks. We should find out
what kind of functions images can play and what cognitive purposes they are better
for. Foremost, what has to be explained is why imagistic thinking strategies are
sometimes better than non-imagistic ones.
2 Historically speaking, there is controversy as to who in fact represents the imagistic
theory of thought. Berkeley appears to believe that all ideas are images (for the
opposite view, see, e.g., Kasem, 1989). Hume believes something similar, though
there are some doubts (e.g. Yolton, 1996). Locke’s view is far from clear. It can be
argued that we do not have to refer to images to understand the nature of Locke’s
concept of idea and that we can instead adopt an adverbial interpretation. See, e.g.,
Ayers (1991), Chappell (1994), Lowe (1995), Soles (1999), White (1990) and Yolton
(1996).
3 In the light of grounding cognition in perception, arguing away the innate
elements of the mind seems to be relatively less important. In the case of Locke,
the point was that empiricism would be more convincing if there were no plausible
 Notes 189

nativist account of knowledge that could challenge his position. In general,

nativism is not incompatible with the imagistic theory of thought. For instance, in
cognitive psychology, conjoining nativism with imagistic theory is instantiated by
the ‘core cognition’ approach. It is based on the assumption that the human mind
may be innately endowed with some basic, presumably iconic, cognitive systems
that explain the capacity to learn exhibited by human infants (e.g. Carey, 2009,
2011; Spelke and Kinzler, 2007) which could help make sense of the apparently
retrospective nature of the representations that underlie many looking-time
experiments in the violation-of-expectancy paradigm (e.g. Baillargeon et al.,
1985).
4 There are two groupings of evidence: behavioural and neurobiological. The
behavioural evidence is based on experimental results indicating that tasks involving
executing mental operations on perceptual or motor representations interfere
with specific perceptual experiences or actions (e.g. Pecher et al., 2003, 2004).
The neurobiological evidence shows that tasks involving mental representations
operations recruit perceptual areas of the human brain (e.g. Beauchamp and Martin,
2007).
5 That does not mean that there are no cognitive effects on perception. For instance,
perceptual interpretation of an ambiguous picture differs before and after
information is given about what the picture is of (e.g. Fazekas and Nanay, 2017;
Teufel and Nanay, 2017). It does not, however, influence the visual mechanism of
picture-perception.
6 The most striking findings come from research on number sense (e.g. Cantlon,
2012; Dehaene, 2011; Izard et al., 2009; Machery, 2007). There is also a number of
neurobiological findings that indicate the amodal nature of concepts (e.g. Bedny
et al., 2008, 2012; Mahon, 2015).
7 The operational approach is not a kind of a functional explanatory strategy, since
it does not determine any causal roles. The explanatory goal is to determine the
necessary conditions of some logical operations.
8 This is a paraphrase of the strategy David Lewis took with respect to explaining
the nature of meaning: ‘In order to say what a meaning is, we may first ask what a
meaning does, and then find something that does that’ (1970, 22).

Chapter 4

1 I hold that knot diagrams are necessary to represent the ways of constructing
knots. This does not mean that every property of a knot is representable by a
knot diagram. For instance, knots have properties represented only by continued
fractions and polynomials.
190 Notes

2 Notably, in the commentary to Euclid’s Elements, Proclus (1992, 159) distinguishes

the concept of construction from the concept of proof. The latter refers to a
process of drawing inferences by means of rigorous reasoning from propositions to
conclusions. Hilbert showed that Euclidean constructions are dispensable in proofs;
however, they were never supposed to be a kind of proofs.
3 Tichy (2012) uses the concept of construction to explicate Frege’s theory of sense
and reference. It is, however, irrelevant for our purposes.
4 Intuitively, it may seem offensive to interpret algebraic formulas as instantiations of
diagrams. Yet, we should not overestimate the power of this intuition. Historically
speaking, the prevailing belief was rather the opposite. For instance, Kant
understands ‘7 + 5’ in terms of a kind of construction, not in terms of a power
of the sets (e.g. Shabel, 2003). According to Kant, algebra ‘achieves by a symbolic
construction equally well what geometry does by an ostensive or geometrical
construction’ (Kant, 1998, A717/B745). The same intuition can be found in Peirce.
He holds that (CP 2.282) ‘every algebraical equation is an icon, in so far as it
exhibits, by means of the algebraic signs (which are not themselves icons), the
relation of the quantities concerned’. According to Peirce, basic units of algebraic
constructions are not iconic. Yet, they do not have to be. What makes algebraic
equations diagrams is that they demonstrate how we can construct magnitudes.
Thus, for Peirce (CP 3.419), ‘algebra is but a sort of diagram’. Let us follow this
tradition and see where it can lead us.
5 The concept of construction is defined very broadly. In general, everything can
be taken as an effect of a construction. Yet, although everything is a product of a
construction, not everything is a representation of a construction. The term ‘7 + 5’
can represent a procedure of arriving at ‘12’, as well as be an effect of a construction
operation representing the proposition 7 + 5 = 12.
6 It may seem natural to interpret the content of diagrams in terms of representing
Platonic ideal objects. In this book, I do not offer a conclusive argument against
Platonism, although I take this view untenable. However, interpreting the content
of diagrams in terms of the rules of construction offers an alternative to Platonism.
Consider the concept of the inflation growth. According to the view presented here,
inflation growth is representable in diagrams. However, inflation growth is not an
ideal object the diagram resembles. The inflation growth is an effect of construction
procedures based on the application of the ordering function to the inflation
rate and time. Does this mean that inflation growth is unreal? No! It means that
inflation growth is not an (ideal) object like a table or chair. However, providing an
argument against Platonism in favour of more credible ontology of mathematical
objects would require a different book.
7 If we think of diagrams as representations that show how to construct relations
within some space, we can abstract from the kind of space we are in, for the
properties of the space do not determine the intrinsic properties of the diagram.
 Notes 191

Thus, the space can be a physical or logical space. For instance, the inflation
diagram localizes the values of time and inflation rate and determines the relation
between them. The term ‘7 + 5 = 12’ determines the relation between some
magnitudes. Geometrical diagrams are only a subset of a more general phenomena.
8 The distinction between type-diagram and token-diagram is Peircean in origin (e.g.
NEM 4.315).
9 Possessing such disposition does not imply that it will be actualized. For instance,
one can have no talent for drawing diagrams. In the same way, if I speak in some
language, I have a disposition to read in this language. However, such disposition
cannot be actualized, if I do not know the alphabet.
10 Note that I claim that diagrams can help discover a proof strategy (e.g. Antonietti,
1991; Giaquinto, 2008; Stenning and Oberlander, 1995). I do not claim that
diagrams can be used as proofs in mathematics. As is well known, we can be easily
misled by accidental properties of diagrams. The concept of mathematical proof is
different from the concept of mathematical construction.
11 As Black once said (1972, 122): ‘My chief objection to the resemblance view, then,
is that when pursued it turns out to be uninformative, offering a trivial verbal
substitution in place of insight.’
12 This structure-preserving mapping does not have to be isomorphism. It can be
either homomorphism (if the mapping is not bijective), monomorphism (if it is
an injective homomorphism), epimorphism (if it is surjective homomorphism),
endomorphism or automorphism (e.g. Pero and Suárez, 2016; Pietarinen, 2014).
Determining the character of these mappings is irrelevant for my argument.
13 Frigg (2006, 50) introduces the so-called problem of style. It concerns the taxonomy
of different styles of scientific representations. The riddle of style is similar to the
problem of the multiple models, which can be expressed by the question of why
scientists use different models of the same targets (Giere, 2006).

Chapter 5

1 Goodman holds (1976) that the idea of naturalness of icons can be explained by the
concept of entrenchment. In a nutshell, icons are natural representations for we are
more familiar with the representational systems that have a longer history of use.
However, that cannot work as an explanation, for we can ask what properties of the
system make it more likely to be entrenched and reproduced in history.
2 Although many philosophers of perception assume that we can perceive states of
affairs (e.g. Armstrong, 1997; Fish, 2009; McDowell, 1996), it is not uncontested (e.g.
Vernazzani, 2021). For our purposes, it is sufficient to claim that if the perception of
facts exists, it is distinct from perceptual recognition.
192 Notes

3 In contrast to Gauker (2011), I do not hold that iconic representations cannot

represent particulars as belonging to kinds. Gauker’s claim seems to be challenged
by the phenomenon of categorical perception. It refers to the common fact that
we perceive distinct categories when there is a gradual change in a variable along
a continuum. For instance, when we see a rainbow we experience colour stripes
despite the stimulus is a smooth gradient of electromagnetic frequencies.
4 Much depends on how we think about the nature of memory. In contrast to the
causal theories (e.g. Bernecker, 2010), the simulation theories (e.g. Michaelian,
2016) hold that memory draws on information acquired during past experience to
construct a simulation of a target event. Memory is understood as a construction
system, rather than a system based on retrieving information. However, regardless
of how we think about the nature of memory, the causal story of how recognition
works seems to be wrong.
5 This general description of recognition-based identification does not determine
the psychological mechanisms of how such invariants are recognized. It can be
reconciled with both view- and structure-based approaches to object recognition in
cognitive psychology (e.g. Hummel, 2013).
6 Of course, it says little about how Cezanne achieves this goal in practice, for
example, how the properties of the picture’s surface, such as brush strokes in a
painting, can produce the experience of seeing-in depicted colours. However, I
strongly doubt that philosophers can answer this question. In the same way, I doubt
that philosophers are in a position to answer the question of how strings of letters
on a piece of paper are transformed into meaningful units by a cognitive system.
Philosophers can ask about the nature of the meaning of language, not about the
psychological mechanisms responsible for grasping the meaning. That is the job of
psycholinguists. By the same token, a philosopher cannot answer the question of
what construction invariants are for representing objects in practice.

Chapter 6

1 Greenberg (2018, 2021), following Cummins, introduces similar distinction

between the target and content. According to Greenberg’s Three-Part Model, the
referent is a particular individual which the representation is of identified by the
image singular content. Singular content is contrasted with the attributive content
which identifies the properties ascribed to those individuals, for example their
shape or orientation. The 2-D model differs from Three-Part Model in at least
three respects. First, according to the 2-D model, a referent does not have to be a
particular individual. Content can identify kinds. Second, the properties identified
by the pictorial content are the construction properties. For the reasons mentioned
 Notes 193

later, I hold that images lack predicative functions. Third, I hold that iconic content
cannot be descriptive.
2 This distinction has been already applied to depiction in phenomenological
tradition (e.g. Husserl, 2005; Ingarden, 1989; Wiesing, 2009). Husserl distinguishes
between the picture’s surface, its content (Bildobjekt) and target (Bildsujet).
3 Sometimes denotation is interpreted as restricted to linguistic representations, such
as proper names. However, this restriction is neither necessary nor helpful (Elgin,
2010b).
4 To deal with targetless representations, Goodman (1976) distinguishes between
representations-of and representations-as. A picture of a unicorn is not a
representation of a unicorn but a kind of a picture, namely, a unicorn-picture.
Kinds of pictures are distinguished by different modes of presentation. I do not
endorse this distinction, for it reintroduces the old problem of sense-reference
distinction. In the same way as there can be no mode of presentation without
something that is represented (Evans, 1982), there can be no representation-as
without representation-of. The same objection applies to Hyman’s (2012, 2015)
distinction between the sense and reference of pictorial representations.
5 Notably, we have to remember to keep aside representing a particular object by
iconic content and identifying the possible world we are in. I hold that iconic
content can identify particulars but is not relativized to the context of possible
worlds (target is).
6 It is important not to confuse the understandability of images with their
interpretability. Understandability of images refers to their ability to be used in
some contexts. Interpretability abstracts from the context of use and determines the
conditions that have to be satisfied by an image to bear a representational function.
An image can be interpretable and not understandable. Suppose that in some
population there are no mathematicians, and there is no one who can understand a
mathematical graph. Still, the graph is interpretable. It does not work the other way
round. An image cannot be understandable if it is not interpretable.
7 This description is complementary with the representational theory of measurement
(RTM). According to RTM, measurement is ‘the construction of homomorphisms
(scales) from empirical relational structures of interest to numerical structures that
are useful’ (Krantz et al., 1971, 9). Measurement should satisfy two kinds of theorems.
The representation theorem holds that there a represented domain can be mapped
onto the structure of the representing domain. The uniqueness theorem establishes
the permissible transformations of the represented domain into the numerical
domain. However, our description of measurement does not depend on RTM.
8 Thinking about images in terms of measurement devices seems to imply that
images possess arbitrary semantics. The argument can look as follows. If we can
measure the world in any way we like, then we can depict the world any way we
want, which is false. Images are naturally connected to the world. We cannot
194 Notes

depict horses as cows, whereas we can conduct measurements with many different
measures. We can measure the length in metres and inches. However, this argument
is invalid, for it is not true that measurements are arbitrarily connected to the
world. We have different measurement conventions in the same way as there are
different iconic conventions. Yet, it is not true that there is no difference in what
type of measurement we choose. We can try to measure distances with gas, but it
does not have the properties that enable identifying the distance. Thus, the way we
measure depends on how the world is. In the same way, images are measurement
devices that are naturally connected to the world. We can depict objects differently.
However, the depiction has to possess properties that identify the object of
depiction. Thus, images depend on what they represent.
9 The extensions of the Reutersvaard-Penrose triangle are the well-known Penrose
stairs or Escher’s ‘Waterfall’.
10 Indeterminable constructions have to be distinguished from indetermined
constructions. For instance, a picture of a man in a hat does not determine whether
he is bald. However, it is a determinable property of this class of constructions. We
can have a different picture of the same man without a hat.
11 The impossible images phenomenon is not restricted to spatial representations.
Indeterminable constructions can be instantiated by temporal representations. For
instance, echolocation jamming in bats occurs when non-target sounds interfere
with target echoes and can be caused by the echolocation system itself.

Chapter 7

1 Note that the concept of understating comes in degrees. One does not have to
understand a whole theory to understand its parts. Most people understand basic
mathematical calculus, but it does not imply that they understand the advanced
calculus.
2 This does not imply that division of language into meaningful parts is semantically
non-arbitrary (Johnson, 2004; Quine, 1970; Salje, 2019). Yet, it is believed that
language has a logical structure and a syntax.
3 Burge (2018) argues that we can establish canonical decomposition by discovering
perceptual competences that determine picture constituents. By the same token, we
can establish a picture’s grammar-like structure. However, his argument misses the
point, for Fodor holds that there is no principal way we can decompose a picture
into semantical parts. Fodor’s argument is metaphysical, not psychological.
4 In Kozak (2021), I distinguish between the so-called vehicle-specificity and
modality-specificity of images.
 Notes 195

5 It appears that it would be hard to understand the concept of mental imagery if it

had nothing in common with the concept of images (e.g. Anderson, 1978).
6 For the discussion on the significance of dissociations between neural mechanisms
of mental imagery and perception, see Bartolomeo et al. (2020) and Pearson (2020).
7 One can ask what mental images taken as conscious experiences are. That is an
important question which, sadly, I cannot answer. It would require having a good
theory of consciousness. This book is not about consciousness.
8 Amodal completion is not a visual phenomenon. It ranges over all modalities. For
instance, we amodally complete a beeped part of a soundtrack (Nanay, 2018).
 Literature

Aasen, S. (2016). ‘Pictures, Presence and Visibility’. Philosophical Studies, 173 (1):
187–203.
Abel, G. (2012). ‘Knowledge Research: Extending and Revising Epistemology’. In A.
Abel and J. Conant (eds), Rethinking Epistemology, Vol. 1, 1–52. Berlin: DeGruyter.
Abell, C. (2009). ‘Canny Resemblance: An Account of Depiction’. Philosophical Review,
118 (2): 183–223.
Abell, C. and K. Bantinaki (2010). ‘Introduction’. In C. Abell and K. Bantinaki (eds),
Philosophical Perspectives on Depiction, 1–24. Oxford; New York: Oxford University
Press.
Abell, C. and G. Currie (1999). ‘Internal and External Pictures’. Philosophical Psychology,
12 (4): 429–45.
Abid, G. (2021). ‘Recognition and the Perception–Cognition Divide’. Mind & Language,
1–20. https://doi​.org​/10​.1111​/mila​.12362.
Abrahamsen, A. and W. Bechtel (2015). ‘Diagrams as Tools for Scientific Reasoning’.
Review of Philosophy and Psychology, 6: 117–31.
Ambrosio, C. (2014). ‘Iconic Representations and Representative Practices’.
International Studies in the Philosophy of Science, 28 (3): 255–75.
Amit, E., C. Hoeflin, N. Hamzah and E. Fedorenko (2017). ‘An Asymmetrical
Relationship Between Verbal and Visual Thinking: Converging Evidence from
Behavior and fMRI’. Neuroimage, 15, 619–27.
Anderson, J. R. (1978). ‘Arguments Concerning Representations for Mental Imagery’.
Psychological Review, 85 (4): 249–77.
Antonietti, A. (1991). ‘Why Does Mental Visualisation Facilitate Problem-Solving?’
In R. H. Logie and M. Denis (eds), Mental Images in Human Cognition, 211–27.
Amsterdam: Elsevier.
Armstrong, D. (1968). A Materialist Theory of the Mind. London: Routledge & Kegan
Paul.
Armstrong, D. (1989). A Combinatorial Theory of Possibility. Cambridge, MA:
Cambridge University Press.
Armstrong, D. (1997). A World of States of Affairs. Cambridge, MA: Cambridge
University Press.
Arnheim, R. (1969). Visual Thinking. Berkeley; Los Angeles: University of California
Press.
Arnheim, R. (1980). ‘A Plea for Visual Thinking’. Critical Inquiry, 6 (3): 489–97.
Ayers, M. (1991). Locke: Epistemology and Ontology. London: Routledge.
 Literature 197

Baddeley, A. D. (1988). ‘Imagery and Working Memory’. In M. Denis, J. Engelkamp and

J. T. E. Richardson (eds), Cognitive and Neuropsychological Approaches to Mental
Imagery, 169–80. Dordrecht: Martinus Nijhoff.
Baddeley, A. D. and K. Lieberman (1980). ‘Spatial Working Memory’. In R. S. Nickerson
(ed.), Attention and Performance, 521–39. Hillsdale, NJ: Erlbaum.
Baillargeon, R., E. Spelke and S. Wasserman (1985). ‘Object Permanence in Five-
Month-Old Infants’. Cognition, 20: 191–208.
Ball, D. N. (2018). ‘Semantics as Measurement’. In D. Ball and B. Rabern (eds), The
Science of Meaning: Essays on the Metatheory of Natural Language Semantics,
381–410. Oxford: Oxford University Press.
Bantinaki, K. (2007). ‘Pictorial Perception as Illusion’. British Journal of Aesthetics, 47
(3): 268–79.
Barsalou, L. W. (1999). ‘Perceptual Symbol Systems (With Commentaries and Author’s
Reply)’. Behavioral and Brain Sciences, 22: 577–660.
Barsalou, L. W. (2010). ‘Grounded Cognition: Past, Present, and Future’. Topics in
Cognitive Science, 2: 716–24.
Bartolomeo, P., D. Hajhajate, J. Liu and A. Spagna (2020). ‘Assessing the Causal Role of
Early Visual Areas in Visual Mental Imagery’. Nature Reviews Neuroscience, 21: 517.
Barwise, J. and J. Etchemendy (1996). ‘Visual Information and Valid Reasoning’. In
G. Allwein and J. Barwise (eds), Logical Reasoning with Diagrams, 3–26. Oxford:
Oxford University Press.
Barwise, J. and A. Shimojima (1995). ‘Surrogate Reasoning’. Cognitive Studies: Bulletin of
Japanese Cognitive Science Society, 4 (2): 7–27.
Bauer, M. I. and P. N. Johnson-Laird (1993). ‘How Diagrams Can Improve Reasoning’.
Psychological Science, 4, 372–8.
Bayne, T. (2013). Thought: A Very Short Introduction. Oxford: Oxford University Press.
Beauchamp, M. S. and A. Martin (2007). ‘Grounding Object Concepts in Perception
and Action: Evidence from fMRI Studies of Tools’. Cortex, 43 (3): 461–8.
Bechtel, W. (2008). Mental Mechanisms: Philosophical Perspectives on Cognitive
Neuroscience. New York: Lawrence Erlbaum Associates.
Bechtel, W. (2017). ‘Diagrammatic Reasoning’. In L. Magnani and T. Bertolotti (eds),
Handbook of Model-Based Science, 605–18. Dordrecht: Springer.
Beck, J. (2015). ‘Analogue Magnitude Representations: A Philosophical Introduction’.
The British Journal for the Philosophy of Science, 66 (4): 829–55.
Bedny, M., A. Caramazza, E. Grossman, A. Pascual-Leone and R. Saxe (2008).
‘Concepts Are More Than Percepts: The Case of Action Verbs’. The Journal of
Neuroscience, 28: 11347–53.
Bedny, M., A. Caramazza, A. Pascual-Leone and R. Saxe (2012). ‘Typical Neural
Representations of Action Verbs Develop Without Vision’. Cerebral Cortex, 22:
286–93.
Beilock, S. L. and S. Goldin-Meadow (2010). ‘Gesture Changes Thought by Grounding
It in Action’. Psychological Science, 21: 1605–10.
198 Literature

Belting, H. (2001). Bild-Anthropologie. Entwürfe für eine Bildwissenschaft. München:

Wilhelm Fink Verlag.
Ben-Yami, H. (1997). ‘Against Characterizing Mental States as Propositional Attitudes’.
The Philosophical Quarterly, 47: 84–9.
Bennett, M. and P. Hacker (2003). Philosophical Foundations of Neuroscience. Oxford:
Blackwell.
Bensafi, M., J. Porter, S. Pouliot, J. Mainland, B. Johnson, C. Zelano, N. Young, E.
Bremner, D. Aframian, R. Kahn and N. Sobel (2003). ‘Olfactomotor Activity
During Imagery Mimics that During Perception’. Nature Neuroscience, 6 (11):
1142–4.
Bensafi, M., S. Pouliot and N. Sobel (2005). ‘Odorant-Specific Patterns of Sniffing
During Imagery Distinguish “Bad” and “Good” Olfactory Imagers’. Chemical Senses,
30 (6): 521–9.
Berman, D. (2008). ‘A Confession of Images’. Philosophical Practice, 3: 255–66.
Bermúdez, J. L. (2000). ‘Naturalized Sense-Data’. Philosophy and Phenomenal Research,
61, 353–74.
Bermúdez, J. L. (2003). Thinking Without Words. Oxford: Oxford University Press.
Bernecker, S. (2010). Memory: A Philosophical Study. Oxford: Oxford University Press.
Black, M. (1972). ‘How Do Pictures Represent?’ In E. H. Gombrich, J. Hochberg, M.
Black (eds), Art, Perception, and Reality, 95–130. Baltimore; London: The John
Hopkins UP.
Blackwell, S. E. (2020). ‘Emotional Mental Imagery’. In A. Abraham (ed.), The
Cambridge Handbook of the Imagination, 241–57. Cambridge: Cambridge University
Press.
Block, N. (1983a). ‘Mental Pictures and Cognitive Science’. The Philosophical Review, 92
(4): 499–541.
Block, N. (1983b). ‘The Photographic Fallacy in the Debate About Mental Imagery’.
Noûs, 17 (4): 651–62.
Blumson, B. (2009). ‘Defining Depiction’. British Journal of Aesthetics, 49 (2): 143–57.
Blumson, B. (2012). ‘Mental Maps’. Philosophy and Phenomenological Research, 85 (2):
413–34.
Blumson, B. (2014). Resemblance and Representation: An Essay in the Philosophy of
Pictures. Cambridge: Open Book Publishers.
Bocanegra, B. R., F. H. Poletiek, B. Ftitache and A. Clark (2019). ‘Intelligent Problem-
Solvers Externalize Cognitive Operations’. Nature Human Behaviour, 3: 136–42.
Boehm, G. (2007). Wie Bilder Sinn erzeugen: Die Macht des Zeigens. Berlin: Berlin
University Press.
Bordwell, D. (2008). Poetics of Cinema. New York: Routledge.
Botterill, G. and P. Carruthers (1999). The Philosophy of Psychology. Cambridge:
Cambridge University Press.
Boumans, M. (2005). How Economists Model the World into Numbers. New York:
Routledge.
 Literature 199

Bradburn, N. M., N. L. Cartwright and J. Fuller (2017). ‘A Theory of Measurement’. In L.

McClimans (ed.), Measurement in Medicine: Philosophical Essays on Assessment and
Evaluation, 73–88. London: Rownan & Littlefield International.
Braddon-Mitchell, D. and F. Jackson (1996). Philosophy of Mind and Cognition. Oxford:
Blackwell.
Braine, M. D. S. and D. P. O’Brien (1998). ‘The Theory of Mental-Propositional Logic:
Description and Illustration’. In M. D. S. Braine and D. P. O’Brien (eds), Mental
Logic, 79–89. Mahwah, NJ: Lawrence Erlbaum Associates.
Briscoe, R. (2016). ‘Depiction, Pictorial Experience, and Vision Science’. Philosophical
Topics, 44 (2): 43–81.
Brown, J. R. (1999). Philosophy of Mathematics: An Introduction to the World of Proofs
and Pictures. London: Routledge.
Buccella, A. (2021). ‘The Problem of Perceptual Invariance’. Synthese, 199 (5–6):
13883–905.
Budd, M. (1995). Values of Art. London: Allen Lane.
Budd, M. (2008). Aesthetic Essays. Oxford: Oxford University Press.
Bueno, O. and S. French (2011). ‘How Theories Represent’. The British Journal for the
Philosophy of Science, 62 (4): 857–94.
Burge, T. (2010). The Origins of Objectivity. Oxford: Oxford University Press.
Burge, T. (2018). ‘Iconic Representation: Maps, Pictures, and Perception’. In S.
Wuppuluri and F. A. Doria (eds), The Map and the Territory: Exploring the
Foundations of Science, Thought and Reality, 79–100. Cham, Switzerland: Springer.
Burke, L. (1952). ‘On the Tunnel Effect’. Quarterly Journal of Experimental Psychology, 4:
121–38.
Byrne, R. M. J. and P. N. Johnson-Laird (1989). ‘Spatial Reasoning’. Journal of Memory
and Language, 28: 564–75.
Camp, E. (2004). ‘The Generality Constraint and Categorial Restrictions’. Philosophical
Quarterly, 54 (215): 209–31.
Camp, E. (2007). ‘Thinking with Maps’. In J. Hawthorne (ed.), Philosophical Perspectives
21: Philosophy of Mind, 145–82. Oxford: Wiley-Blackwell.
Camp, E. (2015). ‘Logical Concepts and Associative Characterizations’. In E. Margolis
and S. Laurence (eds), The Conceptual Mind: New Directions in the Study of Concepts,
591–621. Cambridge, MA: MIT Press.
Camp, E. (2018). ‘Why Maps Are Not Propositional’. In A. Grzankowski and M.
Montague (eds), Non-Propositional Intentionality, 19–45. Oxford: Oxford University
Press.
Candlish, S. (2001). ‘Mental Imagery’. In S. Schroeder (ed.), Wittgenstein and
Contemporary Philosophy of Mind, 107–28. London; New York: Palgrave
MacMillan.
Cantlon, J. F. (2012). ‘Math, Monkeys, and the Developing Brain’. Proceedings of the
National Academy of Sciences, 109 (Supplement 1): 10725–32.
Carey, S. (2009). The Origin of Concepts. New York: Oxford University Press.
200 Literature

Carey, S. (2011). ‘Précis of “The Origin of Concepts”’. The Behavioral and brain sciences,
34 (3): 113–24.
Cartwright, N. (1983). How the Laws of Physics Lie. Oxford: Oxford University Press.
Cartwright, N. and L. Runhardt (2014). ‘Measurement’. In N. Cartwright and E.
Montuschi (eds), Philosophy of Social Science: A New Introduction, 265–87. Oxford:
Oxford University Press.
Carvalho, J. M. (2019). Thinking with Images: An Enactivist Aesthetics. New York;
London: Routledge.
Casati, R. (2015). ‘Object Perception’. In M. Matthen (ed.), The Oxford Handbook of
Philosophy of Perception, 393–404. Oxford: Oxford University Press.
Casati, R. (2018). ‘Two, then Four Modes of Functioning of the Mind: Towards a
Unification of “Dual” Theories of Reasoning and Theories of Cognitive Artefacts’.
In J. M. Zacks and H. A. Taylor (eds), Representations in Mind and World: Essays
inspired by Barbara Tversky, 7–23. New York: Routledge.
Casati, R. and A. Varzi (1999). Parts and Places: The Structures of Spatial Representation.
Cambridge, MA: MIT Press.
Cavedon-Taylor, D. (2021). ‘Untying the Knot: Imagination, Perception and Their
Neural Substrates’. Synthese, 199: 7203–30.
Chambers, D. and D. Reisberg (1985). ‘Can Mental Images be Ambiguous?’ Journal of
Experimental Psychology: Human Perception and Performance, 11 (3): 317–28.
Chang, H. (2004). Inventing Temperature: Measurement and Scientific Progress. Oxford:
Oxford University Press.
Chang, H. and N. Cartwright (2014). ‘Measurement’. In M. Curd and S. Psillos (eds),
The Routledge Companion to Philosophy of Science, 411–19. London; New York:
Routledge.
Chappell, V. (1994). ‘Locke’s Theory of Ideas’. In V. Chappell (ed.), The Cambridge
Companion to Locke, 26–55. Cambridge: Cambridge University Press.
Chevalier, J.-M. (2015). ‘The Problem of Resemblance in Peirce’s Semiotics and
Philosophy’. Versus, 120: 45–59.
Church, J. (2008). ‘The Hidden Image: A Defense of Unconscious Imagining and Its
Importance’. American Imago, 65: 379–404.
Clark, A. (2013). ‘Whatever Next? Predictive Brains, Situated Agents, and the Future of
Cognitive Science’. Behavioral and Brain Sciences, 36: 181–204.
Clement, C. A. and R. J. Falmagne (1986). ‘Logical Reasoning, World Knowledge, and
Mental Imagery: Interconnections in Cognitive Processes’. Memory and Cognition,
14: 299–307.
Collins, C. (1991). The Poetics of the Mind’s Eye: Literature and the Psychology of
Imagination. Philadelphia, PA: University of Pennsylvania Press.
Coopmans, C. (2014). ‘Visual Analytics as Artful Revelation’. In C. Coopmans, J.
Vertesi, M. Lynch and S. Woolgar (eds), Representation in Scientific Practice
Revisited, 37–60. Cambridge, MA; London: MIT Press.
Crane, T. (1988). ‘The Waterfall Illusion’. Analysis, 48: 142–7.
 Literature 201

Crane, T. (2001). Elements of Mind. Oxford: Oxford University Press.

Crane, T. (2009). ‘Is Perception a Propositional Attitude?’ The Philosophical Quarterly,
59: 452–69.
Craven, J. and D. H. Foster (1992). ‘An Operational Approach to Colour Constancy’.
Vision Research, 32 (7): 1359–66.
Croijmans, I., L. Speed, A. Arshamian and A. Majid (2019). ‘Measuring the
Multisensory Imagery of Wine: The Vividness of Wine Imagery Questionnaire’.
Multisensory Research, 32 (3): 179–95.
Cromwell, P. (2004). Knots and Links. Cambridge: Cambridge University Press.
Cummins, R. (1996). Representations, Targets, and Attitudes. Cambridge, MA; London:
MIT Press.
Curry, G. (1995). Image and Mind: Film, Philosophy and Cognitive Science. New York:
Cambridge University Press.
Davidson, D. (1982). ‘Rational Animals’. Dialectica, 36 (4): 317–27.
Davidson, D. (1997). ‘Seeing Through Language’’. In J. Preston (ed.), Thought and
Language, 15–27. Cambridge: Cambridge University Press.
Davidson, D. (2001). Inquiries into Truth and Interpretation. Oxford: Oxford University
Press.
Davies, W. (2016). ‘Colour Constancy, Illumination, and Matching’. Philosophy of
Science, 83: 540–62.
Dawes, A. J., R. Keogh, T. Andrillon and J. Pearson (2020). ‘A Cognitive Profile of Multi-
Sensory Imagery, Memory and Dreaming in Aphantasia’. Scientific Reports, 10 (1):
1–10.
Defeyter, M. A., R. Russo and P. L. McPartlin (2009). ‘The Picture Superiority Effect
in Recognition Memory: A Developmental Study Using the Response Signal
Procedure’. Cognitive Development, 24 (3): 265–73.
Dehaene, S. (2011). The Number Sense: How The Mind Creates Mathematics. Oxford:
Oxford University Press.
DeLoache, J., S. L. Pierroutsakos and D. H. Uttal (2003). ‘The Origins of Pictorial
Competence’. Current Directions in Psychological Science, 12 (4): 114–18.
Denis, M. (1991). ‘Imagery and Thinking’. In C. Cornoldi and M. McCaniel (eds),
Imagery and Cognition, 103–31. New York: Springer Verlag.
Dennett, D. C. (1981a). ‘The Nature of Images and the Introspective Trap’. In N. Block
(ed.), Imagery, 51–62. New York; London: MIT Press.
Dennett, D. C. (1981b). ‘Two Approaches to Mental Images’. In. N. Block (ed.), Imagery,
87–107. New York; London: MIT Press.
Dennett, D. C. (1987). The Intentional Stance. Cambridge, MA: MIT Press.
Dennett, D. C. (1990). ‘Thinking with a Computer’. In H. Barlow, C. Blakemore and M.
Westo-Smith (eds), Images and Understanding: Thoughts about Images: Ideas about
Understanding, 297–309. Cambridge: Cambridge University Press.
Dennett, D. C. (1991a). Consciousness Explained. Boston, MA: Little, Brown.
Dennett, D. C. (1991b). ‘Real Patterns’. Journal of Philosophy, 88 (1): 27–51.
202 Literature

Dennett, D. C. (2013). Intuition Pumps and Other Tools for Thinking. New York: W. W.
Norton & Company.
Dennett, D. C. (2017). From Bacteria to Bach and Back: The Evolution of Minds. New
York: W. W. Norton & Company.
De Regt, H. W. (2014). ‘Visualization as a Tool for Understanding’. Perspectives on
Science, 22 (3): 377–96.
De Regt, H. W. (2017). Understanding Scientific Understanding. New York: Oxford
University Press.
De Toffoli, S. and V. Giardino (2014). ‘Forms and Roles of Diagrams in Knot Theory’.
Erkenntnis, 79 (4): 829–42.
Devitt, M. (2006). Ignorance of Language. Oxford: Oxford University Press.
Dilworth, J. (2008). ‘The Propositional Challenge to Aesthetics’. British Journal of
Aesthetics, 48 (2): 115–44.
Djordjevic, J., R. J. Zatorre, M. Petrides, J. A. Boyle and M. Jones-Gotman (2005).
‘Functional Neuroimaging of Odor Imagery’. Neuroimage, 24: 791–801.
Djordjevic, J., R. J. Zatorre, M. Petrides and M. Jones-Gotman (2004). ‘The Mind’s Nose:
Effects of Odor and Visual Imagery on Odor Detection’. Psychological Science, 15 (3):
143–8.
Dokic, J. (2010). ‘Perceptual Recognition and the Feeling of Presence’. In B. Nanay (ed.),
Perceiving the World, 33–53. Oxford: Oxford University Press.
Dominowski, R. L. and L. E. Bourne (1994). ‘History of Research on Thinking and
Problem Solving’. In R. J. Sternberg (ed.), Thinking and Problem Solving, 1–36. San
Diego, CA: Academic Press.
Dorsch, F. (2016). ‘Knowledge by Imagination – How Imaginative Experience Can
Ground Factual Knowledge’. Teorema: Revista Internacional De Filosofía, 35 (3):
87–116.
Dresner, E. (2002). ‘Measurement Theoretic Semantics and Semantics of Necessity’.
Synthese, 130: 413–40.
Dresner, E. (2006). ‘A Measurement Theoretic Account of Propositions’. Synthese, 153:
1–22.
Dresner, E. (2010). ‘Language and the Measure of Mind’. Mind & Language, 25 (4):
418–39.
Dresner, E. (2014). ‘Decision Theory, Propositional Measurement, and Unified
Interpretation’. Mind, 123 (491): 707–32.
Dretske, F. (1981). Knowledge and the Flow of Information. Cambridge, MA: MIT
Press.
Dummett, M. (1993). Origins of Analytical Philosophy. London: Duckworth.
Earnshaw, R. A. and D. Watson (eds) (1993). Animation and Scientific Visualization:
Tools and Applications. London: Academic Press.
Eaton, M. (1980). ‘Truth in Pictures’. The Journal of Aesthetics and Art Criticism, 39 (1):
15–26.
Eco, U. (1995). The Search for the Perfect Language. Oxford: Blackwell.
 Literature 203

Egan, D. (2011). ‘Pictures in Wittgenstein’s Later Philosophy’. Philosophical

Investigations, 34 (1): 55–76.
Elgin, C. Z. (1983). With Reference to Reference. Indianapolis; Cambridge: Hackett.
Elgin, C. Z. (1996a). Considered Judgment. Princeton, NJ: Princeton University Press.
Elgin, C. Z. (1996b). ‘Index and Icon Revisited’. In V. M. Colapietro and T. M.
Olshewsky (eds), Peirce's Doctrine of Signs: Theory, Applications, and Connections,
181–9. Berlin; New York: de Gruyter.
Elgin, C. Z. (2010a). ‘Keeping Things in Perspective’. Philosophical Studies, 150:
439–47.
Elgin, C. Z. (2010b). ‘Telling Instances’. In R. Frigg and M. C. Hunter (eds), Beyond
Mimesis and Convention: Representation in Art and Science, 1–18. Berlin; New York:
Springer.
Elgin, C. Z. (2017). True Enough. Cambridge, MA; London: MIT Press.
Ellis, R. D. (1995). Questioning Consciousness: The Interplay of Imagery, Cognition, and
Emotion in the Human Brain. Amsterdam: John Benjamins.
Elpidorou, A. (2016). ‘Seeing the Impossible’. Journal of Aesthetics and Art Criticism, 74
(1): 11–21.
Ernst, B. (1986). Adventures with Impossible Figures. Norfolk: Tarquin Publications.
Ertz, T.-P. (2008). Regel und Witz: Wittgensteinsche Perspektiven auf Mathematik,
Sprache und Moral. Berlin; New York: de Gruyter.
Evans, G. (1982). The Varieties of Reference. Oxford: Oxford University Press.
Evans, J. St B. T. (2017). Thinking and Reasoning: A Very Short Introduction. Oxford:
Oxford University Press.
Fazekas, P. and B. Nanay (2017). ‘Pre-Cueing Effects: Attention or Mental Imagery’.
Frontiers in Psychology, 8: 222.
Ferguson, E. S. (1992). Engineering and the Mind’s Eye. Cambridge, MA: MIT Press.
Finke, R. A. (1989). Principles of Mental Imagery. Cambridge, MA: MIT Press.
Finke, R. A., S. Pinker and M. J. Farah (1989). ‘Reinterpreting Visual Patterns in Mental
Imagery’. Cognitive Science, 13 (1): 51–78.
Finkelstein, L. (2003). ‘Widely, Strongly and Weakly Defined Measurement’.
Measurement, 34: 39–48.
Firestone, C. and B. J. Scholl (2016). ‘Cognition Does Not Affect Perception: Evaluating
the Evidence for “Top-Down” Effects’. Behavioral and Brain Sciences, 36: 1–77.
Fish, W. (2009). Perception, Hallucination, and Illusion. New York: Oxford University
Press.
Fodor, J. (1975). The Language of Thought. Cambridge, MA: Harvard University Press.
Fodor, J. (1987). Psychosemantics: The Problem of Meaning in the Philosophy of Mind.
Cambridge, MA: MIT Press.
Fodor, J. (1991). ‘Replies’. In B. Loewer and G. Rey (eds), Meaning in Mind: Fodor and
his Critics, 255–319. Oxford: Basil Blackwell.
Fodor, J. (2007). ‘The Revenge of the Given’. In B. P. McLaughlin and J. D. Cohen (eds),
Contemporary Debates in Philosophy of Mind, 105–16. Oxford: Basil Blackwell.
204 Literature

Fodor, J. (2008). LOT 2: The Language of Thought Revisited. Oxford: Oxford University
Press.
Fodor, J. and Z. Pylyshyn (1981). ‘How Direct is Visual Perception?: Some Reflections
on Gibson’s “Ecological Approach”’. Cognition, 9: 207–46.
Fodor, J. and Z. Pylyshyn (1988). ‘Connectionism and Cognitive Architecture: A
Critical Analysis’. Cognition, 28 (1–2): 3–71.
Foster, D. (2003). ‘Does Colour Constancy Exist?’ Trends in Cognitive Sciences, 7: 439–43.
Foster, D. (2011). ‘Colour Constancy’. Vision Research, 51: 674–700.
Francis, G. K. (2007). A Topological Picturebook. New York: Springer Verlag.
Frege, G. (1984). ‘Thoughts’. Translated by P. Geach and R. Stoothoff. In B. McGuinness
(ed.), Collected Papers on Mathematics, Logic, and Philosophy, 351–72. Oxford: Basil
Blackwell.
French, S. R. D. (2003). ‘A Model-Theoretic Account of Representation (Or, I Don't
Know Much About Art But I Know It Involves Isomorphism)’. Philosophy of Science,
70 (5): 1472–83.
Frigg, R. (2006). ‘Scientific Representation and the Semantic View of Theories’. Theoria,
21 (1): 49–65.
Frigg, R and J. Nguyen (2020). Modelling Nature: An Opinionated Introduction to
Scientific Representation. Cham: Springer.
Frixione, M. and A. Lombardii (2015). ‘Street Signs and Ikea Instruction Sheets: Pragmatics
and Pictorial Communication’. Review of Philosophy and Psychology, 6: 133–49.
Galton, F. (1880). ‘Statistics of Mental Imagery’. Mind, 5: 301–18.
Gardner, H. (2004). Frames of Mind: The Theory of Multiple Intelligences. New York:
Basic Books.
Gaskin, R. (2008). The Unity of the Proposition. Oxford: Oxford University Press.
Gauker, C. (2007). ‘A Critique of the Similarity Space Theory of Concepts’. Mind &
Language, 22 (4): 317–45.
Gauker, C. (2011). Words and Images: An Essay on the Origin of Ideas. Oxford: Oxford
university Press.
Gauker, C. (2017). ‘Three Kinds of Nonconceptual Seeing-as’. Review of Philosophy and
Psychology, 8 (4): 763–79.
Gauker, C. (2020). ‘On the Difference Between Realistic and Fantastic Imagining’.
Erkenntnis, 1–20.
Gendler, T. S. (2004). ‘Thought Experiments Rethought – and Reperceived’. Philosophy
of Science, 71: 1154–63.
Gersel, J. P. (2017). ‘Discriminatory Capacities, Russell's Principle, and the Importance
of Losing Sight of Objects’. European Journal of Philosophy, 25: 700–20.
Giaquinto, M. (2007). Visual Thinking in Mathematics: An Epistemological Study.
Oxford: Oxford University Press.
Giaquinto, M. (2008). ‘Visualizing in Mathematics’. In P. Mancosu (ed.), The Philosophy
of Mathematical Practice, 22–42. Oxford: Oxford University Press.
 Literature 205

Giaquinto, M. (2011). ‘Crossing Curves: A Limit to the Use of Diagrams in Proofs’.

Philosophia Mathematica, 19: 281–307.
Giardino, V. (2014). ‘Diagramming: Connecting Cognitive Systems to Improve
Reasoning’. In A. Benedek and K. Nyiri (eds), The Power of the Image: Emotion,
Expression, Explanation, 23–34. Frankfurt/M: Peter Lang.
Giardino, V. (2016). ‘Behind the Diagrams: Cognitive Issues and Open Problems’. In
S. Krämer and C. Ljungberg (eds), Thinking with Diagrams: The Semiotic Basis of
Human Cognition, 77–101. Boston; Berlin: Walter de Gruyter.
Giardino, V. (2017). ‘Diagrammatic Reasoning in Mathematics’. In L. Magnani
and T. Bertolotti (eds), Handbook of Model-Based Science, 499–522. Dordrecht:
Springer.
Giardino, V. and G. Greenberg (2015). ‘Introduction: Varieties of Iconicity’. Review of
Philosophy and Psychology, 6 (1): 1–25.
Gibson, J. J. (1954). ‘A Theory of Pictorial Perception’. Audio Visual Communication
Review, 2 (1): 3–23.
Gibson, J. J. (1971). ‘The Information Available in Pictures’. Leonardo, 4: 27–35.
Gibson, J. J. (1973). ‘On the Concept of “Formless Invariants” in Visual Perception’.
Leonardo, 6: 43–5.
Gibson, J. J. (1978). ‘The Ecological Approach to the Visual Perception of Pictures’.
Leonardo, 11: 227–35.
Gibson, J. J. (1979). The Ecological Approach to the Visual Perception. Houghton, MI:
Mifflin.
Giere, R. N. (2006). Scientific Perspectivism. Chicago; London: University of Chicago
Press.
Gilbert, A. N., M. Crouch and S. E. Kemp (1998). ‘Olfactory and Visual Mental
Imagery’. Journal of Mental Imagery, 22: 137–46.
Giovannelli, A. (2001). ‘Picture, Image and Experience: A Philosophical Inquiry’. Mind,
110: 481–5.
Goldberg, B. (2017). ‘Mechanism and Meaning’. In J. Hyman (ed.), Investigating
Psychology: Sciences of the Mind After Wittgenstein, 48–66. London, New York:
Routledge.
Goldin-Meadow, S. (2003). Hearing Gesture: How Our Hands Help Us to Think.
Cambridge, MA: Harvard University Press.
Goldstone, R. L. and L. W. Barsalou (1998). ‘Reuniting Perception and Conception’.
Cognition, 65: 231–62.
Gombrich, E. H. (1960). Art and Illusion: A Study in the Psychology of Pictorial
Representation. London: Phaidon.
Gombrich, E. H. (1971). ‘Exchange of Letters’. Leonardo, 4: 195–203.
Gombrich, E. H. (1972). ‘The “What” and the “How”: Perspective Representation and
the Phenomenal World’. In R. S. Rudner and I. Scheffler (eds), Logic and Art: Essays
in Honor of Nelson Goodman, 129–49. Indianapolis, IN: Bobbs-Merrill.
206 Literature

Gombrich, E. H. (1990). ‘Pictorial Instructions’. In H. Barlow, C. Blakemore and M.

Weston-Smith (eds), Images and Understanding, 26–45. Cambridge, MA: Cambridge
University Press.
Gooding, D. C. (1990). Experiment and the Making of Meaning: Human Agency in
Scientific Observation and Experiment. Dordrecht; Boston: Kluwer Academic
Publishers.
Gooding, D. C. (2010). ‘Visualizing Scientific Inference’. Topics in Cognitive Science, 2:
15–35.
Goodman, N. (1972). ‘Seven Strictures on Similarity’. In N. Goodman (ed.), Problems
and Projects, 437–46. Indianapolis; New York: The Bobbs-Merrill Company.
Goodman, N. (1976). Languages of Art, 2nd edn. Indianapolis, IN: Hackett.
Goodman, N. (1984). Of Mind and Other Matters. Cambridge, MA; London: Harvard
University Press.
Goodman, N. (1988). ‘On What Should Not Be Said about Representation’. Journal of
Aesthetics and Art Criticism, 46 (3): 419.
Goodman, N. (1990). ‘Pictures in the Mind?’ In H. Barlow, C. Blakemore and M.
Westo-Smith (eds), Images and Understanding: Thoughts about Images: Ideas about
Understanding, 358–64. Cambridge: Cambridge University Press.
Goodman, N. and C. Z. Elgin (1988). Reconceptions in Philosophy and other Arts and
Sciences. Indianapolis; Cambridge: Hackett.
Goodwin, W. (2009). ‘Visual Representation in Science’. Philosophy of Science, 76:
372–90.
Grandin, T. (2006). Thinking in Pictures and Other Reports from My Life with Autism.
London: Bloomsbury.
Green, E. J. (2019). ‘On the Perception of Structure’. Noûs, 53 (3): 564–92.
Green, E. J. (2020). ‘The Perception-Cognition Border: A Case for Architectural
Division’. Philosophical Review, 129: 323–93.
Green, E. J. and J. Quilty-Dunn (2021). ‘What is an Object File?’ The British Journal for
the Philosophy of Science, 72 (3): 665–99.
Greenberg, D. L., M. J. Eacott, D. Brechin and D. C. Rubin (2005). ‘Visual Memory
Loss and Autobiographical Amnesia: A Case Study’. Neuropsychologia, 43 (10):
1493–502.
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.
Greenberg, G. (2021). ‘Semantics of Pictorial Space’. Review of Philosophy and
Psychology, 12: 847–87.
Gregory, D. (2013). Showing, Sensing, and Seeming: Distinctively Sensory Representations
and their Contents. Oxford: Oxford University Press.
Gregory, D. (2020). ‘Pictures, Propositions, and Predicates’. American Philosophical
Quarterly, 57 (2): 155–70.
Gregory, R. L. (1966). Eye and Brain. London: World University Library.
 Literature 207

Grzankowski, A. (2013). ‘Non-Propositional Attitudes’. Philosophy Compass, 8 (12):

1123–37.
Grzankowski, A. (2015). ‘Pictures Have Propositional Content’. Review of Philosophy
and Psychology, 6: 151–63.
Grzankowski, A. (2018). ‘A Theory of Non-Propositional Attitudes’. In A. Grzankowski
and M. Montague (eds), Non-Propositional Intentionality, 134–51. Oxford: Oxford
University Press.
Guillot, A. (2020). ‘Neurophysiological Foundations and Practical Applications of
Motor Imagery’. In A. Abraham (eds), The Cambridge Handbook of the Imagination,
207–26. Cambridge: Cambridge University Press.
Gurr, C., J. Lee and K. Stenning (1998). ‘Theories of Diagrammatic Reasoning:
Distinguishing Component Problems’. Minds and Machines, 8: 533–57.
Halpern, A. R., R. J. Zatorre, M. Bouffard and J. A. Johnson (2004). ‘Behavioral and
Neural Correlates of Perceived and Imagined Musical Timbre’. Neuropsychologia, 42
(9): 1281–92.
Haugeland, J. (1998). Having Thought: Essays in the Metaphysics of Mind. Cambridge,
MA; London: Harvard University Press.
Hawthorne, J. and D. Manley (2012). The Reference Book. Oxford: Oxford University
Press.
Heck, R. G. (2000). ‘Nonconceptual Content and the “Space of Reasons”’. The
Philosophical Review, 109: 483–523.
Heck, R. G. (2007). ‘Are There Different Kinds of Content?’ In B. P. McLaughlin and
J. Cohen (eds), Contemporary Debates in Philosophy of Mind, 117–38. Oxford:
Blackwell.
Hegarty, M. (1992). ‘Mental Animation: Inferring Motion From Static Displays of
Mechanical Systems’. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 18: 1084–102.
Hegarty, M. and M. Kozhevnikov (1999). ‘Types of Visual-Spatial Representations
and Mathematical Problem Solving’. Journal of Educational Psychology, 91 (4):
684–9.
Hintikka, J. (1987). ‘Mental Models, Semantical Games, and Varieties of Intelligence’.
In L. Vaina (ed.), Matters of Intelligence: Conceptual Structures in Cognitive
Neuroscience, 197–215. Dordrecht: D. Reidel.
Hinton, G. (1979). ‘Some Demonstrations of the Effects of Structural Descriptions in
Mental Imagery’. Cognitive Science, 3 (3): 231–50.
Holyoak, K. J. and R. G. Morrison (2012). ‘Thinking and Reasoning: A Reader’s Guide’.
In K. J. Holyoak and R. G. Morrison (eds), Oxford Handbook of Thinking and
Reasoning, 1–7. New York: Oxford University.
Holyoak, K. J. and P. Thagard (1996). Mental Leaps: Analogy in Creative Thought.
Cambridge, MA; London: MIT Press.
Hookway, C. (2000). Truth, Rationality and Pragmatism. Oxford: Oxford University
Press.
208 Literature

Hookway, C. (2007). ‘Peirce on Icons and Cognition’. In U. Priss, S. Polovina and R. Hill
(eds), Conceptual Structures: Knowledge Architectures for Smart Applications, 59–68.
Berlin; Heidelberg: Springer.
Hope, V. (2009). ‘Object Perception, Perceptual Recognition, and That-Perception
Introduction’. Philosophy, 84 (330): 515–28.
Hopkins, R. (1998). Picture, Image and Experience: A Philosophical Inquiry. Cambridge:
Cambridge University Press.
Hubbard, T. L. (1997). ‘Target Size and Displacement Along the Axis of Implied
Gravitational Attraction: Effects of Implied Weight and Evidence of Representational
Gravity’. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23:
1484–93.
Hubbard, T. L. (2010). ‘Auditory Imagery: Empirical Findings’. Psychological Bulletin,
136 (2): 302–29.
Hume, D. (1975). A Treatise of Human Nature. Ed. L. A. Selby-Bigge, 2nd edn., revised
by P. H. Nidditch. Oxford: Clarendon Press.
Hume, D. (1977). An Enquiry Concerning Human Understanding. Ed. E. Steinberg.
Indianapolis, IN: Hackett.
Hummel, J. E. (2013). ‘Object Recognition’. In D. Reisberg (ed.), Oxford Handbook of
Cognitive Psychology, 32–46. Oxford: Oxford University Press.
Husserl, E. (2005). Phantasy, Image Consciousness, and Memory (1898–1925). Trans. R.
Bernet. Dordrecht: Springer.
Hyman, J. (2006). The Objective Eye: Color, Form, and Reality in the Theory of Art.
Chicago, IL: University of Chicago Press.
Hyman, J. (2012). ‘Depiction’. Royal Institute of Philosophy Supplements, 71: 129–50.
Hyman, J. (2015). ‘Depiction’. In P. F. Bundgaard and F. Stjernfelt (eds), Investigations in
the Phenomenology and the Ontology of the Work of Art: What are Artworks and How
Do We Experience Them?, 191–208. Heidelberg: Springer.
Ingarden, R. (1989). The Ontology of the Work of Art. Trans. R. Meyer and J. T.
Goldthwait. Athens, OH: Ohio University Press.
Inkpin, A. (2016). ‘Projection, Recognition, and Pictorial Diversity’. Theoria, 82:
32–55.
International Organization for Standardization (1994). Accuracy (Trueness and
Precision) of Measurement Methods and Results – Part 1: General Principles and
Definitions. Standard No. 5725-1.
Intons-Peterson, M. J. (1983). ‘Imagery Paradigms: How Vulnerable are They to
Experimenter's Expectations?’ Journal of Experimental Psychology: Human
Perception and Performance, 9: 394–412.
Intons-Peterson, M. J. and A. R. White (1981). ‘Experimenter Naiveté and Imaginal
Judgments’. Journal of Experimental Psychology: Human Perception and Performance,
7 (4): 833–43.
Irvine, E. (2014). ‘What Iconicity Can and Cannot Do for Proto-Language’. In E. A.
Cartmill, S. Roberts, H. Lyn and H. Cornish (eds), The Evolution of Language:
 Literature 209

Proceedings of the 10th International Conference (EVOLANG10), Vienna, Austria,

14–17 April 2014, 122–9. Singapore: World Scientific.
Isaac, A. M. C. (2019). ‘The Allegory of Isomorphism’. AVANT, X (3): 1–23.
Ishai, A., L. G. Ungerleider and J. V. Haxby (2000). ‘Distributed Neural Systems for the
Generation of Visual Images’. Neuron, 28: 979–90.
Ison, M. J. and R. Q. Quiroga (2008). ‘Selectivity and Invariance for Visual Object
Perception’. Frontiers in Bioscience, 13: 4889–903.
Ittelson, W. H. (1996). ‘Visual Perception of Markings’. Psychonomic Bulletin & Review,
3 (2): 171–87.
Izard, V., C. Sann, E. S. Spelke and A. Streri (2009). ‘Newborn Infants Perceive
Abstract Numbers’. Proceedings of the National Academy of Sciences, 106 (25):
10382–5.
Jakubowski, K. (2020). ‘Musical Imagery’. In A. Abraham (ed.), The Cambridge
Handbook of the Imagination, 187–206. Cambridge: Cambridge University Press.
Jäkel, F., M. Singh, F. A. Wichmann and M. H. Herzog (2016). ‘An Overview of
Quantitative Approaches in Gestalt Perception’. Vision Research, 126: 3–8.
Johnson, K. (2004). ‘On the Systematicity of Language and Thought’. Journal of
Philosophy, 101 (3): 111–39.
Johnson-Laird, P. N. (1998). ‘Imagery, Visualization, and Thinking’. In J. Hochberg (ed.),
Perception and Cognition at the Century's End, 441–67. San Diego, CA: Academic
Press.
Kamermans, K. L., W. Pouw, F. W. Mast and F. Paas (2019). ‘Reinterpretation in Visual
Imagery Is Possible Without Visual Cues: A Validation of Previous Research’.
Psychological Research: An International Journal of Perception, Attention, Memory
and Action, 83 (6): 1237–50.
Kan, I. P., L. W. Barsalou, K. O. Solomon, J. K. Minor and S. L. Thompson-Schill
(2003). ‘Role of Mental Imagery in a Property Verification Task: fMRI Evidence for
Perceptual Representations of Conceptual Knowledge’. Cognitive Neuropsychology,
20: 525–40.
Kant, I. (1998). Critique of Pure Reason. Trans. P. Guyer and A. Wood. Cambridge:
Cambridge University Press.
Kaplan, D. (1968). ‘Quantifying’. Synthese, 19 (1–2): 178–214.
Kaplan, D. (1989). ‘Demonstratives’. In J. Almog, J. Perry, and H. Wettstein (eds),
Themes from Kaplan, 481–563. New York: Oxford University Press.
Kasem, A. (1989). ‘Can Berkeley Be Called an Imagist?’ Indian Philosophical Quarterly,
6: 75–88.
Kaufmann, G. (1996). ‘The Many Faces of Mental Imagery’. In C. Cornoldi, R. H. Logie,
M. A. Brandimonte, G. Kaufmann and D. Reisberg (eds), Stretching the Imagination:
Representation and Transformation in Mental Imagery, 77–118. New York: Oxford
University Press.
Kennedy, J. M. (1974). A Psychology of Picture Perception. San Francisco; Washington;
London: Jossey-Bass Publishers.
210 Literature

Keogh, R. and J. Pearson (2018). ‘The Blind Mind: No Sensory Visual Imagery in
Aphantasia’. Cortex, 105: 53–60.
Khalifa, K. (2017). Understanding, Explanation, and Scientific Knowledge. New York:
Cambridge University Press.
Kiefer, M. and F. Pulvermüller (2012). ‘Conceptual Representations in Mind and Brain:
Theoretical Developments, Current Evidence and Future Directions’. Cortex, 48: 805–25.
Kind, A. (2018). ‘How Imagination Gives Rise to Knowledge’. In F. Macpherson and F.
Dorsch (eds), Perceptual Memory and Perceptual Imagination, 227–46. New York:
Oxford University Press.
Kirsh, D. (2010). ‘Thinking with External Representations’. Artificial Intelligence and the
Simulation of Behaviour, 25: 441–54.
Kirsh, D. and P. Maglio (1994). ‘On Distinguishing Epistemic from Pragmatic Action’.
Cognitive Science, 18: 513–49.
Kitcher, P. and A. Varzi (2000). ‘Some Pictures are Worth 2[aleph]0 Sentences’.
Philosophy, 75 (3): 377–81.
Kjørup, S. (1974). ‘George Inness and the Battle at Hastings, or Doing Things With
Pictures’. Monist, 58 (2): 216–35.
Klatzky, R. L., S. J. Lederman and D. E. Matula (1991). ‘Imagined Haptic Exploration
in Judgements of Object Properties’. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 17: 314–22.
Knauff, M. (2013). Space to Reason: A Spatial Theory of Human Thought. Cambridge,
MA; London: MIT Press.
Knauff, M. and P. N. Johnson-Laird (2002). ‘Visual Imagery Can Impede Reasoning’.
Memory and Cognition, 30: 363–71.
Kobayashi, M., M. Takeda, N. Hattori, M. Fukunaga, T. Sasabe, N. Inoue, Y. Nagai, T.
Sawada, N. Sadato and Y. Watanabe (2004). ‘Functional Imaging of Gustatory Perception
and Imagery: “Top-Down” Processing of Gustatory Signals’. NeuroImage, 23: 1271–82.
Kosslyn S. M. (1980). Image and Mind. Cambridge, MA: Harvard University Press.
Kosslyn, S. M. (1994). Image and Brain: The Resolution of the Imagery Debate.
Cambridge, MA: MIT Press.
Kosslyn, S. M., T. M. Ball and B. J. Reiser (1978). ‘Visual Images Preserve Metric Spatial
Information: Evidence from Studies of Image Scanning’. Journal of Experimental
Psychology: Human Perception and Performance, 4: 47–60.
Kosslyn, S. M., W. L. Thompson and G. Ganis (2006). The Case for Mental Imagery.
Oxford: Oxford University Press.
Kosso, P. (2007). ‘Scientific Understanding’. Foundations of Science, 12: 173–88.
Kozak, P. (2020) ‘The Diagram Problem’. In A.-V. Pietarinen, P. Chapman, L. Bosveld-de
Smet, V. Giardino, J. Corter and S. Linker (eds), Diagrammatic Representation and
Inference, 216–24. Cham: Springer.
Kozak, P. (2021). ‘The Analog-Digital Distinction Fails to Explain the Perception-
Thought Distinction: An Alternative Account of the Format of Mental
Representation’. Studia Semiotyczne, 35 (1): 73–94.
 Literature 211

Kozhevnikov, M., S. Kosslyn and J. Shephard (2005). ‘Spatial vs. Object Visualizers: A New
Characterization of Visual Cognitive Style’. Memory and Cognition, 33 (4): 710–26.
Kozhevnikov, M., O. Blazhenkova and M. Becker (2010). ‘Trade-Off in Object Versus
Spatial Visualization Abilities: Restriction in the Development of Visual-Processing
Resources’. Psychonomic Bulletin & Review, 17 (1): 29–35.
Krantz, D., R. D. Luce, A. Tversky and P. Suppes (1971). Foundations of Measurement
Volume I: Additive and Polynomial Representations. San Diego and London:
Academic Press.
Kripke, S. A. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press.
Kripke, S. A. (1982). Wittgenstein on Rules and Private Language. Oxford: Blackwell.
Kulpa, Z. (1987). ‘Putting Order in the Impossible’. Perception, 16: 201–14.
Kulpa, Z. (2009). ‘Main Problems of Diagrammatic Reasoning. Part I: The
Generalization Problem’. Foundations of Science, 14 (1–2): 75–96.
Kulvicki, J. (2006a). On Images: Their Structure and Content. Oxford: Clarendon Press.
Kulvicki, J. (2006b). ‘Pictorial Representation’. Philosophy Compass, 1 (6): 535–46.
Kulvicki, J. (2014). Images. New York: Routledge.
Kulvicki, J. (2020). Modeling the Meanings of Pictures: Depiction and the Philosophy of
Language. Oxford: Oxford University Press.
Kvanvig, J. L. (2018). ‘Knowledge, Understanding, and Reasons for Belief ’. In D. Starr
(ed.), The Oxford Handbook of Reasons and Normativity, 685–715. New York: Oxford
University Press.
Lakoff, G. (1994). ‘What is Metaphor?’ In J. A. Barnden and K. J. Holyoak (eds),
Advances in Connectionist and Neural Computation Theory: Vol. 3, Analogy,
Metaphor, and Reminding, 203–57. Norwood, NJ: Ablex.
Lakoff, G. and M. Johnson (1980). Metaphors We Live By. Chicago, IL: University of
Chicago Press.
Langacker, R. W. (1987). Foundations of Cognitive Grammar. Stanford, CA: Stanford
University Press.
Langland-Hassan, P. (2015). ‘Imaginative Attitudes’. Philosophy and Phenomenological
Research, 40: 664–86.
Langland-Hassan, P. (2020). Explaining Imagination. Oxford: Oxford University Press.
Larkin, J. H. and H. A. Simon (1987). ‘Why a Diagram is (Sometimes) Worth Ten
Thousand Words’. Cognitive Science, 11: 65–99.
Latour, B. (1990). ‘Visualisation and Cognition: Drawing Things Together’. In M. Lynch
and S. Woolgar (eds), Representation in Scientific Activity, 19–68. Cambridge, MA:
MIT Press.
Laurence, S. and E. Margolis (2012). ‘The Scope of the Conceptual’. In E. Margolis, R.
Samuels and S. Stich (eds), The Oxford Handbook of Philosophy of Cognitive Science,
291–317. New York: Oxford University Press.
Lehrer, K. (2000). ‘Meaning, Exemplarization and Metarepresentation’. In D. Sperber
(ed.), Metarepresentations: A Multidisciplinary Perspective, 299–310. Oxford: Oxford
University Press.
212 Literature

Lemon, O., M. de Rijke and A. Shimojima (1999). ‘Efficacy of Diagrammatic Reasoning’.

Journal of Logic, Language, and Information, 8 (3): 265–71.
Lewis, D. (1970). ‘General Semantics’. Synthese, 22: 18–67.
Lewis, D. (1971). ‘Analog and Digital’. Noûs, 5: 321–8.
Lindsay, R. K. (1998). ‘Using Diagrams to Understand Geometry’. Computational
Intelligence, 14 (2): 222–56.
Lipton, P. (2009). ‘Understanding Without Explanation’. In H. de Regt, S. Leonelli,
and K. Eigner (eds), Scientific Understanding: Philosophical Perspectives, 43–63.
Pittsburgh, PA: University of Pittsburgh.
List, C. J. (1981). ‘Images, Propositions, and Natural Signs’. In E. Klinger (ed.), Concepts,
Results, and Applications, 67–76. Boston, MA: Springer.
Liu, Z., D. C. Knill and D. Kersten (1995). ‘Object Classification for Human & Ideal
Observers’. Vision Research, 35: 549–68.
Locke, J. (1975). An Essay Concerning Human Understanding. Ed. P. H. Nidditch.
Oxford: Oxford University Press.
Loewenstein, J. and D. Gentner (2005). ‘Relational Language and the Development of
Relational Mapping’. Cognitive Psychology, 50 (4): 315–53.
Lopes, D. M. (1996). Understanding Pictures. Oxford: Oxford University Press.
Lopes, D. M. (1997). ‘Art Media and the Sense Modalities: Tactile Pictures’. The
Philosophical Quarterly, 47: 425–40.
Lopes, D. M. (2003). ‘Pictures and the Representational Mind’. The Monist, 86 (4): 632–52.
Lopes, D. M. (2005). Sight and Sensibility: Evaluating Pictures. Oxford: Oxford
University Press.
Lopes, D. M. (2010). ‘Picture This: Image-Based Demonstratives’. In C. Abell and K.
Bantinaki (eds), Philosophical Perspectives on Depiction, 52–80. Oxford; New York:
Oxford University Press.
Lowe, D. G. (2004). ‘Distinctive Image Features From Scale-Invariant Keypoints’.
International Journal of Computer Vision, 60 (2): 91–110.
Lowe, E. J. (1995). Locke on Human Understanding. London: Routledge.
Lowe, E. J. (1996). Subjects of Experience. Cambridge: Cambridge University Press.
Lupyan, G., D. H. Rakison and J. L. McClelland (2007). ‘Language Is Not Just for
Talking: Redundant Labels Facilitate Learning of Novel Categories’. Psychological
Science, 18 (12): 1077–83.
Macbeth, D. (2010). ‘Diagrammatic Reasoning in Euclid’s Elements’. In B. Van
Kerkhove, J. De Vuyst and J. P. Van Bendegem (eds), Philosophical Perspectives on
Mathematical Practice, 235–67. London: College Publications.
Macbeth, D. (2012). ‘Seeing How It Goes: Paper-and-Pencil Reasoning in Mathematical
Practice’. Philosophia Mathematica, 20: 58–85.
Machery, E. (2007). ‘Concept Empiricism: A Methodological Critique’. Cognition, 104:
19–46.
Machery, E. (2016). ‘The Amodal Brain and the Offloading Hypothesis’. Psychonomic
Bulletin & Review, 23: 1090–5.
 Literature 213

MacKisack, M., S. Aldworth, F. Macpherson, J. Onians, C. Winlove and A. Zeman

(2016). ‘On Picturing a Candle: The Prehistory of Imagery Science’. Frontiers in
Psychology, 7: 515.
Mahon, B. Z. (2015). ‘What is Embodied About Cognition?’ Language, Cognition and
Neuroscience, 30: 420–9.
Maley, C. J. (2011). ‘Analog and Digital, Continuous and Discrete’. Philosophical Studies,
155: 117–13.
Malinas, G. (1991). ‘A Semantics for Pictures’. Canadian Journal of Philosophy, 21 (3):
275–98.
Mancosu, P. (2005). ‘Visualization in Logic and Mathematics’. In P. Mancosu, K.
Jørgensen and S. Pedersen (eds), Visualization, Explanation and Reasoning Styles in
Mathematics, 13–30. Dordrecht: Springer.
Manders, K. (2008). ‘The Euclidean Diagram’. In P. Mancosu (ed.), The Philosophy of
Mathematical Practice, 80–133. Oxford: Oxford University Press.
Marcus, R. B. (1990). ‘Some Revisionary Proposals About Belief and Believing’.
Philosophy and Phenomenological Research, 50: 132–153.
Martin, A. (2007). ‘The Representation of Object Concepts in the Brain’. Annual Review
of Psychology, 58: 25–45.
Mast, F. W. and S. M. Kosslyn (2002). ‘Visual Mental Images Can Be Ambiguous:
Insights From Individual Differences in Spatial Transformation Abilities’. Cognition,
86: 57–70.
Matthen, M. (2005). Seeing, Doing, and Knowing: A Philosophical Theory of Sense
Perception. Oxford: Clarendon Press.
Matthen, M. (2014). ‘Image Content’. In B. Brogaard (ed.), Does Perception Have
Content?, 265–90. New York: Oxford University Press.
Matthews, R. J. (2007). The Measure of Mind: Propositional Attitudes and their
Attribution. Oxford: Oxford University Press.
Matthews, R. J. (2011). ‘Measurement-Theoretic Accounts of Propositional Attitudes’.
Philosophy Compass, 6 (11): 828–41.
Maynard, P. (2011). ‘What Drawing Draws On: The Relevance of Current Vision
Research’. Rivista di estetica, 47 (2): 9–29.
Maynard, P. (2017). ‘Photo Mensura’. In N. Mößner and A. Nordmann (eds), Reasoning
in Measurement, 41–56. London; New York: Routledge.
McCarthy, R. A. and E. K. Warrington (1986). ‘Visual Associative Agnosia: A Clinico-
Anatomical Study of a Single Case’. Journal of Neurology Neurosurgery & Psychiatry,
49 (11): 1233–40.
McDowell, J. (1996). Mind and World. Cambridge, MA: Harvard University Press.
McGinn, C. (1989). Mental Content. Oxford: Blackwell Publishers.
McGinn, C. (2006). Mindsight: Image, Dream, Meaning. Cambridge, MA: Harvard
University Press.
McNeill, D. (1992). Hand in Mind: What Gestures Reveal about Thought. Chicago;
London: Chicago University Press.
214 Literature

Mersch, D. (2003). ‘Wort, Bild, Ton, Zahl: Medialitäten Medialen Darstellens’. In D.

Mersch (ed.), Die Medien der Künste, 9–49. München: Wilhelm Fink Verlag.
Mersch, D. (2011). ‘Aspects of Visual Epistemology: On the “Logic” of the Iconic’. In
A. Benedek and K. Nyiri (eds), Images in Language: Metaphors and Metamorphoses,
169–94. Frankfurt a. M.: Peter Lang.
Merricks, T. (2009). ‘Propositional Attitudes?’. Proceedings of the Aristotelian Society,
109: 207–32.
Meynell, L. (2018). ‘Images and Imagination in Thought Experiments’. In M. Stuart, Y.
Fehige and J. R. Brown (eds), The Routledge Companion to Thought Experiments,
498–511. New York: Routledge.
Meynell, L. (2020). ‘Getting the Picture: Towards a New Account of Scientific
Understanding’. In M. Ivanova and S. French (eds), The Aesthetics of Science: Beauty,
Imagination and Understanding, 36–62. New York: Routledge.
Michaelian K. (2016). Mental Time Travel: Episodic Memory and Our Knowledge of the
Personal Past. Cambridge, MA: MIT Press.
Mitchell, W. J. T. (1994). Picture Theory: Essays on Verbal and Visual Representation.
Chicago; London: University of Chicago Press.
Molitor, S., S. Ballstaedt and H. Mandl (1989). ‘Problems in Knowledge Acquisition
From Text and Pictures’. In H. Mandl and J. R. Levin (eds), Knowledge Acquisition
from Text and Pictures, 3–30. Amsterdam: North-Holland.
Montague, M. (2007). ‘Against Propositionalism’. Noûs, 41 (3): 503–18.
Morrison, M. and M. S. Morgan (1999). ‘Models as Mediating Instruments’. In M. S.
Morgan and M. Morrison (eds), Models as Mediators: Perspectives on Natural and
Social Science, 10–38. Cambridge: Cambridge University Press.
Mortensen, C. (2010). Inconsistent Geometry. London: College Publications.
Mortensen, C., S. Leishman, P. Quigley and T. Helke (2013). ‘How Many Impossible
Images Did Escher Produce?’ British Journal of Aesthetics, 53 (4): 425–41.
Mößner, N. (2018). Visual Representations in Science: Concept and Epistemology.
London; New York: Routledge.
Mullarkey, J. (2011). ‘Temple Grandin’s Animal Thoughts: On Non-Human Thinking
in Pictures, Films, and Diagrams’. In A. Benedek and K. Nyiri (eds), Images in
Language: Metaphors and Metamorphoses, 155–68. Frankfurt a. M.: Peter
Lang.
Mumma, J. (2010). ‘Proofs, Pictures, and Euclid’. Synthese, 175 (2): 255–87.
Murez, M. and F. Recanati (2016). ‘Mental Files: An Introduction’. Review of Philosophy
and Psychology, 7: 265–81.
Müller, A. (1997). Die ikonische Differenz: Das Kunstwerk als Augenblick. München:
Wilhelm Fink Verlag.
Nail, T. (2019). Theory of the Image. New York: Oxford University Press.
Nanay, B. (2010). ‘Perception and Imagination: Amodal Perception as Mental Imagery’.
Philosophical Studies, 150: 239–54.
Nanay, B. (2014). Aesthetics as Philosophy of Perception. Oxford: Oxford University Press.
 Literature 215

Nanay, B. (2015). ‘Perceptual Content and the Content of Mental Imagery’. Philosophical
Studies, 172: 1723–36.
Nanay, B. (2017). ‘Pain and Mental Imagery’. The Monist, 100: 485–500.
Nanay, B. (2018). ‘The Importance of Amodal Completion in Everyday Perception’.
i-Perception, 9 (4): 1–16.
Nanay, B. (2021). ‘Unconscious Mental Imagery’. Philosophical Transactions of the Royal
Society B, 376 (1817): 20190689.
Narens, L. (1985). Abstract Measurement Theory. Cambridge, MA: MIT Press.
Neander, K. (1987). ‘Pictorial Representation: A Matter of Resemblance’. British Journal
of Aesthetics, 27 (3): 213–26.
Nelsen, R. B. (1993). Proofs Without Words: Exercises in Visual Thinking. Washington:
The Mathematical Association of America.
Nersessian, N. (2008). Creating Scientific Concepts. Cambridge, MA: MIT Press.
Newell, A. and H. A. Simon (1972). Human Problem Solving. Englewood Cliffs, NJ:
Prentice Hall.
Newall, M. (2011). What is a Picture? Depiction, Realism, Abstraction. London: Palgrave
Macmillan.
Newton, N. (1982). ‘Experience and Imagery’. The Southern Journal of Philosophy, 21:
475–87.
Novitz, D. (1977). Pictures and Their Use in communication: A Philosophical Essay. The
Hague: Martinus Nijhoff.
Novitz, D. (1988). ‘Review of Deeper into Pictures: An Essay on Pictorial
Representation’. British Journal of Aesthetics, 28 (1): 87–9.
Nunberg, G. (1993). ‘Indexicality and Deixis’. Linguistics and Philosophy, 16: 1–43.
Nyíri, J. C. (2001). ‘The Picture Theory of Reason’. In B. Brogaard and B. Smith (eds),
Rationality and Irrationality (Schriftenreihe-Wittgenstein Gesellschaft, Vol. 29),
242–66. Vienna: Öbv&hpt.
Nyíri, K. (2013). ‘Wittgenstein’s Philosophy of Pictures’. In A. Pichler and S. Säätelä
(eds), Wittgenstein: The Philosopher and his Works, 322–53. Berlin; Boston: De
Gruyter.
Otis, L. (2015). Rethinking Thought: Inside the Minds of Creative Scientists and Artists.
Oxford; New York: Oxford University Press.
Overgaard, S. (2022). ‘Seeing-as, Seeing-o, and Seeing-that’. Philosophical Studies, 179
(11): 1–20.
Pagin, P. and D. Westerståhl (2010). ‘Compositionality I: Definitions and Variants’.
Philosophy Compass, 5: 250–64.
Paivio, A. (1963). ‘Learning of Adjective-Noun Paired Associates as a Function
of Adjective-Noun Word Order and Noun Abstractness’. Canadian Journal of
Psychology, 17: 370–79.
Paivio, A. and R. Harshman (1983). ‘Factor Analysis of a Questionnaire on Imagery and
Verbal Habits and Skills’. Canadian Journal of Psychology, 37: 461–83.
Peacocke, C. (1986). Thoughts: An Essay on Content. Oxford: Basil Blackwell.
216 Literature

Peacocke, C. (1987). ‘Depiction’. Philosophical Review, 96 (3): 383–410.

Peacocke, C. (1992). A Study of Concepts. Cambridge, MA: MIT Press.
Peacocke, C. (2019). The Primacy of Metaphysics. Oxford: Oxford University Press.
Pearson, J. (2020). ‘Reply to: Assessing the Causal Role of Early Visual Areas in Visual
Mental Imagery’. Nature Reviews Neuroscience, 21: 517.
Pearson, J. and S. M. Kosslyn (2015). ‘The Heterogeneity of Mental Representation:
Ending the Mental Imagery Debate’. Proceedings of the National Academy of Sciences
PNAS (PNAS), 11: 10089–92.
Pearson, J., T. Naselaris, E. A. Holmes and S. M. Kosslyn (2015). ‘Mental Imagery:
Functional Mechanisms and Clinical Applications’. Trends in Cognitive Sciences, 19
(10): 590–602.
Pecher, D., R. Zeelenberg and L. W. Barsalou (2003). ‘Verifying Different-Modality
Properties for Concepts Produces Switching Costs’. Psychological Science, 14: 119–24.
Pecher, D., R. Zeelenberg and L. W. Barsalou (2004). ‘Sensorimotor Simulations
Underlie Conceptual Representations: Modality-Specific Effect of Prior Act’.
Psychonomic Bulletin & Review, 11: 164–7.
Peirce, C. S. (1931–1935). Collected Papers of Charles Sanders Peirce, Vol. 1–6. Ed. C.
Hartshorne and P. Weiss. Cambridge, MA: Harvard University Press (Cited as CP n.
m where n and m indicate volume and paragraph numbers).
Peirce, C. S. (1976). The New Elements of Mathematics, Vol. 1–4. Ed. C. Eisele. Hague:
Mouton Publishers; Atlantic Highlands, NJ: Humanities Press (Cited as NEM n. m
where n and m indicate volume and page numbers).
Perini, L. (2005). ‘The Truth in Pictures’. Philosophy of Science, 72: 262–85.
Perini, L. (2012). ‘Truth-Bearers or Truth-Makers?’ Spontaneous Generations: A Journal
for the History and Philosophy of Science, 6 (1): 142–7.
Pero, F. and M. Suárez (2016). ‘Varieties of Misrepresentation and Homomorphism’.
European Journal for Philosophy of Science, 6 (1): 71–90.
Perry, J. (1979). ‘The Problem of the Essential Indexical’. Noûs, 13 (1): 3–21.
Peterson, M. A., J. F. Kihlstrom, P. M. Rose and M. L. Glisky (1992). ‘Mental Images
Can Be Ambiguous: Reconstruals and Reference Frame Reversals’. Memory and
Cognition, 20: 107–23.
Pietarinen, A.-V. (2006). Signs of Logic: Peircean Themes on the Philosophy of Language,
Games, and Communication. Dordrecht: Springer.
Pietarinen, A.-V. (2014). ‘A Road to the Philosophy of Iconic Communication’. In
A. Benedek and K. Nyiri (eds), The Power of the Image: Emotion, Expression,
Explanation, 47–60. Frankfurt a. M.: Peter Lang.
Pietarinen, A.-V. (2016). ‘Is There a General Diagram Concept?’ In S. Krämer and C.
Ljungberg (eds), Thinking with Diagrams: The Semiotic Basis of Human Cognition,
121–37. Boston; Berlin: Walter de Gruyter.
Podro, M. (1998). Depiction. London: Yale University Press.
Price, H. H. (1953). Thinking and Experience. London: Hutchinson’s University
Library.
 Literature 217

Priest, G. (1997). ‘Sylvan’s Box: A Short Story and Ten Morals’. Notre Dame Journal of
Formal Logic, 38 (4): 573–82.
Priest, G. (1999). ‘Perceiving Contradictions’. Australasian Journal of Philosophy, 77 (4):
439–46.
Prinz, J. J. (2002). Furnishing the Mind: Concepts and Their Perceptual Basis. Cambridge,
MA: MIT Press.
Proclus (1992). A Commentary on the First Book of Euclid’s Elements. Trans. G. Morrow.
Princeton, NJ: Princeton University Press.
Putnam, H. (1981). Reason, Truth, and History. Cambridge: Cambridge University
Press.
Pylyshyn, Z. W. (1973). ‘What the Mind’s Eye Tells the Mind’s Brain: A Critique of
Mental Imagery’. Psychological Bulletin, 80: 1–25.
Pylyshyn, Z. W. (2002). ‘Mental Imagery: In Search of a Theory’. Behavioral and Brain
Sciences, 25: 157–82.
Pylyshyn, Z. W. (2003a). Seeing and Visualizing: It’s Not What You Think. Cambridge,
MA: MIT Press.
Pylyshyn, Z. W. (2003b). ‘Return of the Mental Image: Are There Really Pictures in the
Brain?’ TRENDS in Cognitive Sciences, 7 (3): 113–18.
Pylyshyn, Z. W. (2007). Things and Places: How the Mind Connects with the World.
Cambridge, MA: MIT Press.
Quilty-Dunn, J. (2016). ‘Iconicity and the Format of Perception’. Journal of
Consciousness Studies, 23 (3–4): 255–63.
Quilty-Dunn, J. (2020). ‘Perceptual Pluralism’. Noûs, 54 (4): 807–38.
Quilty-Dunn, J. and E. Mandelbaum (2020). ‘Non-Inferential Transitions: Imagery and
Association’. In T. Chan and A. Nes (eds), Inference and Consciousness, 151–71. New
York: Routledge.
Quine, W. V. O. (1956). ‘Quantifiers and Propositional Attitudes’. Journal of Philosophy,
5: 177–87.
Quine, W. V. O. (1960). Word and Object. Cambridge, MA: MIT Press.
Quine, W. V. O. (1968). ‘Ontological Relativity’. Journal of Philosophy, 65: 185–212.
Quine, W. V. O. (1970). ‘Methodological Reflections on Current Linguistic Theory’.
Synthese, 21 (3–4): 386–98.
Raftopoulos, A. (2009). Cognition and Perception: How do Psychology and Neural
Science Inform Philosophy? Cambridge, MA: MIT Press.
Raftopoulos, A. and J. Zeimbekis (2015). ‘The Cognitive Penetrability of Perception –
An Overview’. In J. Zembeiskis and A. Raftopolous (eds), The Cognitive Penetrability
of Perception: New Philosophical Perspectives, 1–56. Oxford: Oxford University
Press.
Ragni, M. and M. Knauff (2013). ‘A Theory and a Computational Model of Spatial
Reasoning with Preferred Mental Models’. Psychological Review, 120 (3): 561–88.
Rayo, A. (2013). The Construction of Logical Space. Oxford: Oxford University Press.
Recanati, F. (2012). Mental Files. New York: Oxford University Press.
218 Literature

Reisberg, D. (1996). ‘The Nonambiguity of Mental Images’. In C. Cornoldi, R. H. Logie,

M. A. Brandimonte, G. Kaufmann and D. Reisberg (eds), Stretching the Imagination:
Representation and Transformation in Mental Imagery, 119–72. New York: Oxford
University Press.
Reisberg, D. and D. Chambers (1991). ‘Neither Pictures Nor Propositions: What We
Can Learn From a Mental Image?’ Canadian Journal of Psychology, 45: 336–52.
Reisberg, D. and F. Heuer (2005). ‘Visuospatial Images’. In P. Shah and A. Miyake (eds),
The Cambridge Handbook of Visuospatial Thinking, 35–80. New York: Cambridge
University Press.
Reisberg, D., D. G. Pearson and S. M. Kosslyn (2003). ‘Intuitions and Introspections
About Imagery: The Role of Imagery Experience in Shaping an Investigator’s
Theoretical Views’. Applied Cognitive Psychology, 17: 147–60.
Rescorla, M. (2009a). ‘Cognitive Maps and the Language of Thought’. British Journal for
the Philosophy of Science, 60: 377–407.
Rescorla, M. (2009b). ‘Predication and Cartographic Representation’. Synthese, 169 (1):
175–200.
Rey, G. (1995). ‘A Not “Merely Empirical” Argument for the Language of Thought’.
Philosophical Perspectives, 9: 201–22.
Rheingold, H. R. (1991). Virtual Reality. London: Secker and Warburg.
Richardson, J. T. E. (1987). ‘The Role of Mental Imagery in Models of Transitive
Inference’. British Journal of Psychology, 78: 189–203.
Richler, J. J., J. B. Wilmer and I. Gauthier (2017). ‘General Object Recognition is
Specific: Evidence from Novel and Familiar Objects’. Cognition, 166: 42–55.
Richtmeyer, U. (2019). Wittgensteins Bilddenken. Leiden, Niederlande: Wilhelm Fink.
Rocke, A. J. (2010). Image and Reality: Kekulé, Kopp, and the Scientific Imagination.
Chicago, IL: University of Chicago Press.
Rollins, M. (1989). Mental Imagery: On the Limits of Cognitive Science. New Haven, CT:
Yale University Press.
Roser, A. (1996). ‘Gibt es Autonome Bilder? Bemerkungen zum Grafischen Werk Otto
Neuraths und Ludwig Wittgensteins’. Grazer Philosophische Studien, 52 (1):
9–43.
Rozemond, M. (1993). ‘Evans on de re Thought’. Philosophia, 22 (3–4): 275–98.
Russell, B. (1919). ‘On Propositions: What They Are and How They Mean’. Aristotelian
Society Supplementary Volume, 2: 1–43.
Russell, B. (1921). The Analysis of Mind. London: Allen & Unwin.
Russell, B. (1997). The Problems of Philosophy. Oxford: Oxford University Press.
Ryle, G. (1949). The Concept of Mind. London: Hutchinson.
Ryle, G. (1979). On Thinking. Oxford: Basil Blackwell.
Ryle, G. (2009). Collected Essays 1929–1968: Collected Papers, Vol. 2. Ed. J. Tanney.
London; New York: Routledge.
Sainsbury, R. M. (2005). Reference Without Referents. Oxford: Oxford University
Press.
 Literature 219

Salis, F. and R. Frigg (2020). Capturing the Scientific Imagination. In A. Levy and P.
Godfrey-Smith (eds), The Scientific Imagination: Philosophical and Psychological
Perspectives, 17–50. New York: Oxford University Press.
Salje, L. (2019). ‘Talking Our Way to Systematicity’. Philosophical Studies, 176: 2563–88.
Sartre, J.-P. (1962). Imagination: A Psychological Critique. Ann Arbor, MI: University of
Michigan Press.
Sartwell, C. (1991). ‘Natural Generativity and Imitation’. British Journal of Aesthetics, 31
(1): 58–67.
Scaife, M. and Y. Rogers (1996). ‘External Cognition: How Do Graphical
Representations Work?’ International Journal of Human-Computer Studies, 45:
185–213.
Scarry, E. (1999). Dreaming by the Book. Princeton, NJ: Princeton University Press.
Schier, F. (1986). Deeper Into Pictures. Cambridge: Cambridge University Press.
Schooler, J. W., M. Fallshore and S. M. Fiore (1995). ‘Epilogue: Putting Insight Into
Perspective’. In R. J. Sternberg and J. E. Davidson (eds), The Nature of Insight,
559–88. Cambridge, MA; London: MIT Press.
Schöttler, T. (2017). ‘Pictorial Evidence: On the Rightness of Pictures’. In N. Mößner
and A. Nordmann (eds), Reasoning in Measurement, 111–29. London; New York:
Routledge.
Sebeok, T. A. (1979). The Sign & Its Masters. Austin: University of Texas Press.
Sellars, W. (1949). ‘Language, Rules and Behavior’. In S. Hook (ed.), John Dewey:
Philosopher of Science and Freedom, 289–315. New York: Dial Press.
Sellars, W. (1963). Science, Perception and Reality. Atascadero: Ridgeview.
Sellars, W. (1969). ‘Language as Thought and as Communication’. Philosophy and
Phenomenological Research, 29 (4): 506–27.
Sellars, W. (2007). In the Space of Reasons: Selected Essays of Wilfrid Sellars. Ed. K.
Scharp, R. B. Brandom. Cambridge, MA: Harvard University Press.
Shabel, L. (2003). Mathematics in Kant’s Critical Philosophy: Reflections on Mathematical
Practice. New York: Routledge.
Shah, P. and A. Miyake (1996). ‘The Separability of Working Memory Resources for
Spatial Thinking and Language Processing’. Journal of Experimental Psychology:
General, 125: 4–27.
Shaver, P., L. Pierson and S. Lang (1975). ‘Converging Evidence for the Functional
Significance of Imagery in Problem Solving’. Cognition, 3: 359–75.
Shepard, R. N. (1978). ‘Externalization of Mental Images and the Act of Creation’.
In B. S. Randhawa and W. E. Coffman (eds), Visual Learning, Thinking, and
Communication, 133–89. New York: Academic Press.
Shepard, R. N. (1990). ‘On Understanding Mental Images’. In H. Barlow, C. Blakemore
and M. Westo-Smith (eds), Images and Understanding: Thoughts about Images: Ideas
about Understanding, 365–70. Cambridge: Cambridge University Press.
Shepard, R. N. and L. A. Cooper (1982). Mental Images and Their Transformations.
Cambridge, MA: MIT Press.
220 Literature

Shepard, R. N. and J. Metzler (1971). ‘Mental Rotation of Three-Dimensional Objects’.

Science, 171: 701–3.
Sheredos, B., D. Burnston, A. Abrahamsen and W. Bechtel (2013). ‘Why Do Biologists
Use So Many Diagrams?’ Philosophy of Science, 80 (5): 931–44.
Sheredos, B. and W. Bechtel (2020). ‘Imagining Mechanisms with Diagrams’. In A.
Levy and P. Godfrey-Smith (eds), The Scientific Imagination: Philosophical and
Psychological Perspectives, 178–219. New York: Oxford University Press.
Shimojima, A. (1996). ‘Operational Constraints in Diagrammatic Reasoning’. In J.
Barwise and G. Allwein (eds), Logical Reasoning with Diagrams, 27–48. Oxford:
Oxford University Press.
Shimojima, A. (2001). ‘The Graphic-Linguistic Distinction: Exploring Alternatives’.
In: A. Blackwell (ed.), Thinking with Diagrams, 5–27. Dordrecht: Kluwer Academic
Publishers.
Shimojima, A. (2015). Semantic Properties of Diagrams and Their Cognitive Potentials.
Stanford, CA: CSLI Publications.
Shin, S.-J. (1994). The Logical Status of Diagrams. New York: Cambridge University
Press.
Shin, S.-J. (2002). The Iconic Logic of Peirce’s Graphs. Cambridge, MA: MIT Press.
Shin, S.-J.(2015). ‘The Mystery of Deduction and Diagrammatic Aspects of
Representation’. Review of Philosophy and Psychology, 6 (1): 49–67.
Shin, S.-J. (2016). ‘The Role of Diagrams in Abductive Reasoning’. In S. Krämer and C.
Ljungberg (eds), Thinking with Diagrams: The Semiotic Basis of Human Cognition,
57–76. Boston; Berlin: Walter de Gruyter.
Short, T. S. (2007). Peirce's Theory of Signs. Cambridge: Cambridge University Press.
Siewert, C. (1998). The Significance of Consciousness. Princeton, NJ: Princeton
University Press.
Skinner, B. F. (1957). Verbal Behavior. New York: Appleton.
Slezak, P. (1995). ‘The “Philosophical” Case Against Visual Imagery’. In P. Slezak, T.
Caelli and R. Clark (eds), Perspectives on Cognitive Science: Theories, Experiments
and Foundations, 237–71. Norwood, NJ: Ablex Publishing.
Slezak, P. (2002a). ‘The Imagery Debate: Déjà vu All Over Again?’ Behavioral and Brain
Sciences, 25 (2): 209–10.
Slezak, P. (2002b). ‘Thinking About Thinking: Language, Thought and Introspection’.
Language & Communication, 22: 353–73.
Sloman, A. (1978). The Computer Revolution in Philosophy: Philosophy, Science, and
Models of Mind. New Jersey: Humanities Press.
Sloman, A. (2002). ‘Diagrams in the Mind?’ In M. Anderson, B. Meyer and P. Olivier
(eds), Diagrammatic Representation and Reasoning, 7–28. London: Springer-Verlag.
Sober, E. (1976). ‘Mental Representations’. Synthese, 33: 101–48.
Soles, D. (1999). ‘Is Locke an Imagist?’ The Locke Newsletter, 30: 17–66.
Solomon, K. O., D. L. Medin and E. B. Lynch (1999). ‘Concepts Do More Than
Categorize’. Trends in Cognitive Sciences, 3: 99–105.
 Literature 221

Solt, K. (1989). ‘Pictures and Truth’. British Journal of Aesthetics, 29 (2): 154–9.
Sorensen, R. (2002). ‘The Art of the Impossible’. In T. S. Gendler and J. Hawthorne
(eds), Conceivability and Possibility, 337–68. Oxford: Oxford University Press.
Spelke, E. S. and K. D. Kinzler (2007). ‘Core Knowledge’. Developmental Science, 10:
89–96.
Squires, R. (1969). ‘Depicting’. Philosophy, 44: 193–204.
Stanley, J. and T. Williamson (2001). ‘Knowing How’. The Journal of Philosophy, 98 (8):
411–44.
Stegmüller, W. (1969). Wissenschaftliche Erklärung und Begründung. Berlin: Springer
Verlag.
Stenning, K. (2002). Seeing Reason: Image and Language in Learning to Think. Oxford:
Oxford University Press.
Stenning, K. and O. Lemon (2001). ‘Aligning Logical and Psychological Perspectives on
Diagrammatic Reasoning’. Artificial Intelligence Review, 15 (1–2): 29–62.
Stenning, K. and J. Oberlander (1995). ‘A Cognitive Theory of Graphical and Linguistic
Reasoning: Logic and Implementation’. Cognitive Science, 19: 97–140.
Stenning, K. and P. Yule (1997). ‘Image and Language in Human Reasoning: A
Syllogistic Illustration’. Cognitive Psychology, 34 (2): 109–59.
Sternberg, R. J. (1980). ‘Representation and Process in Linear Syllogistic Reasoning’.
Journal of Experimental Psychology: General, 109 (2): 119–59.
Stich, S. P. (1978). ‘Beliefs and Subdoxastic States’. Philosophy of Science, 45 (4): 499–518.
Stjernfelt, F. (2000). ‘Diagrams as Centerpiece of a Peircean Epistemology’. Transactions
of the Charles S. Peirce Society, 36 (3): 357–84.
Stjernfelt, F. (2007). Diagrammatology: An Investigation on the Borderlines of
Phenomenology, Ontology and Semiotics. Dordrecht: Springer.
Strawson, P. F. (1963). Individuals: An Essay in Descriptive Metaphysics. New York:
Anchor Books.
Suárez, M. (2003). ‘Scientific Representation: Against Similarity and Isomorphism’.
International Studies in the Philosophy of Science, 17 (3): 225–44.
Swoyer, C. (1987). ‘The Metaphysics of Measurement’. In J. Forge (ed.), Measurement,
Realism and Objectivity: Essays on Measurement in the Social and Physical Sciences,
235–90. Dordrecht: Springer.
Swoyer, C. (1991). ‘Structural Representation and Surrogative Reasoning’. Synthese, 87
(3): 449–508.
Tal, E. (2013). ‘Old and New Problems in Philosophy of Measurement’. Philosophy
Compass, 8 (12): 1159–73.
Tal, E. (2017). ‘A Model-Based Epistemology of Measurement’. In N. Mößner and
A. Nordmann (eds), Reasoning in Measurement, 233–53. London; New York:
Routledge.
Tarski, A. (1956). Logic, Semantics, Metamathematics: Papers from 1923 to 1938. Oxford:
Oxford University Press.
Teller, P. (2013). ‘The Concept of Measurement-Precision’. Synthese, 190: 189–202.
222 Literature

Teller, P. (2018). ‘Measurement Accuracy Realism’. In I. Peschard and B. C. van Fraassen

(eds), The Experimental Side of Modeling, 273–98. Minneapolis, MN: University of
Minnesota Press.
Terrone, E. (2021a). ‘Seeing-in and Singling Out: How to Reconcile Pictures with
Singular Thought’. Pacific Philosophical Quarterly, 102 (3): 378–92.
Terrone, E. (2021b). ‘The Standard of Correctness and the Ontology of Depiction’.
American Philosophical Quarterly, 58 (4): 399–412.
Teufel, C. and B. Nanay (2017). ‘How to (And How Not To) Think About Top-Down
Influences on Perception’. Consciousness and Cognition, 47: 17–25.
Thagard, P. (2005). Mind: Introduction to Cognitive Science. Cambridge, MA; London:
MIT Press.
Thagard, P. (2012). The Cognitive Science of Science: Explanation, Discovery, and
Conceptual Change. Cambridge, MA; London: MIT Press.
Thielen, J., S. E. Bosch, T. M. van Leeuwen, M. van Gerven and R. van Lier (2019).
‘Neuroimaging Findings on Amodal Completion: A Review’. i-Perception, 10 (2):
1–25.
Thomas, N. J. T. (1997). ‘A Stimulus to the Imagination’. PSYCHE: An Interdisciplinary
Journal of Research On Consciousness, 3 (4): 1–14.
Thomas, N. J. T. (2003). ‘Mental Imagery, Philosophical Issues About’. In L. Nadel (ed.),
Encyclopedia of Cognitive Science, Vol. 2, 147–1153. London: Nature Publishing/
Macmillan.
Thomas, N. J. T. (2009). ‘Visual Imagery and Consciousness’. In W. P. Banks (ed.),
Encyclopedia of Consciousness, Vol. 2, 445–57. Oxford: Academic Press.
Tichy, P. (1986). ‘Constructions’. Philosophy of Science, 53: 514–34.
Tichy, P. (2012). The Foundations of Frege's Logic. Berlin; Boston: De Gruyter.
Troscianko, E. T. (2013). ‘Reading Imaginatively: The Imagination in Cognitive Science
and Cognitive Literary Studies’. Journal of Literary Semantics, 42 (2): 181–98.
Tufte, E. R. (1997). Visual Explanations: Images and Quantities, Evidence and Narrative.
Cheshire, CN: Graphics Press.
Tversky, B. (2004). ‘Semantics, Syntax, and Pragmatics of Graphics’. In K. Holmqvist
and Y. Ericsson (eds), Language and Visualization, 141–58. Lund: Lund University
Press.
Tversky, B. (2011). ‘Visualizing Thought’. Topics in Cognitive Science, 3: 499–535.
Tversky, B. (2015). ‘The Cognitive Design of Tools of Thought’. Review of Philosophy and
Psychology, 6: 99–116.
Twardowski, K. (1965). ‘Wyobrażenia i pojęcia’. In K. Twardowski (ed.), Wybrane Pisma
Filozoficzne, 114–97. Warszawa: PWN.
Twardowski, K. (1999). On Actions, Products and Other Topics in Philosophy. Ed. J.
Brandl and J. Wolenski. Amsterdam: Rodopi.
Tye, M. (2005). ‘Non-Conceptual Content, Richness, and Fineness of Grain’. In T.
Gendler and J. Hawthorne (eds), Perceptual Experience, 504–26. Oxford: Oxford
University Press.
 Literature 223

Tye, M. (2012). ‘Knowing What It Is Like’. In J. Bengson and M. A. Moffett (eds),

Knowing How: Essays on Knowledge, Mind, and Action, 300–13. Oxford: Oxford
University Press.
Vallée-Tourangeau, F., S. V. Steffensen, G. Vallée-Tourangeau and M. Sirota (2016).
‘Insight with Hands and Things’. Acta Psychologica, 170: 195–205.
Van Fraassen, B. C. (2008). Scientific Representation: Paradoxes of Perspective. Oxford:
Oxford University Press.
Vernazzani, A. (2021). ‘Do We See Facts?’ Mind & Language, 37 (4): 674–93.
Viera, G. and Nanay, B. (2020). ‘Temporal Mental Imagery’. In A. Abraham (ed.), The
Cambridge Handbook of the Imagination, 227–40. Cambridge: Cambridge University
Press.
Voltolini, A. (2015). A Syncretistic Theory of Depiction. New York: Palgrave Macmillan.
Von Neumann, J. (1958). The Computer and the Brain. New Haven, CT: Yale University
Press.
Walton, K. L. (1973). ‘Pictures and Make-Believe’. The Philosophical Review, 82 (3):
283–319.
Walton, K. L. (1984). ‘Transparent Pictures: On the Nature of Photographic Realism’.
Critical Inquiry, 11 (2): 246–77.
Walton, K. L. (1990). Mimesis as Make-Believe. Cambridge, MA: Harvard University
Press.
Ware, C. (2000). Information Visualization: Perception for Design. San Francisco, CA:
Morgan Kaufmann.
Warrington, E. K. and M. James (1988). ‘Visual Apperceptive Agnosia: A Clinico-
Anatomical Study of Three Cases’. Cortex: A Journal Devoted to the Study of the
Nervous System and Behavior, 24 (1): 13–32.
Watson, J. B. (1913). ‘Psychology as the Behaviorist Views It’. Psychological Review, 20:
158–77.
Westerhoff, J. (2005). ‘Logical Relations between Pictures’. Journal of Philosophy, 102
(12): 603–23.
White, A. R. (1990). The Language of Imagination. Oxford: Blackwell.
Wiesing, L. (2009). Artificial Presence: Philosophical Studies in Image Theory. Trans. N. F.
Schott. Stanford, CA: Stanford University Press.
Willats, J. (1997). Art and Representation: New Principles in the Analysis of Pictures.
Princeton, NJ: Princeton University Press.
Williamson, T. (2002). Knowledge and Its Limits. Oxford: Oxford University Press.
Williamson, T. (2016). ‘Knowing by Imagining’. In A. Kind and P. Kung (eds),
Knowledge Through Imagination, 113–23. New York: Oxford University Press.
Winn, W. (1993). ‘An Account of How Readers Search for Information in Diagrams’.
Contemporary Educational Psychology, 18: 162–85.
Wittgenstein, L. (1953). Philosophical Investigations. Ed. G. E. M. Anscombe and R.
Rhees, trans. G. E. M. Anscombe, 2nd edn., 1958. Oxford: Blackwell.
Wittgenstein, L. (1958). The Blue and Brown Books [1933–1935]. Oxford: Blackwell.
224 Literature

Wollheim, R. (1980). Art and its Objects: Second Edition with Six Supplementary Essays.
Cambridge: Cambridge University Press.
Wollheim, R. (1984). The Thread of Life. Cambridge: Cambridge University Press.
Wollheim, R. (1987). Painting as an Art. Princeton, NJ: Princeton University Press.
Wollheim, R. (1993). The Mind and its Depths. Cambridge, MA; London: Harvard
University Press.
Wollheim, R. (1998). ‘On Pictorial Representation’. The Journal of Aesthetics and Art
Criticism, 56 (3): 217–26.
Wollheim, R. (2003). ‘What Makes Representational Painting Truly Visual?’ Aristotelian
Society Supplementary Volume, 77 (1): 131–47.
Yolton, J. W. (1996). Perception and Reality: A History from Descartes to Kant. Ithaca,
NY: Cornell University Press.
Yoo, S.-S., D. K. Freeman, J. J. III McCarthy and F. A. Jolesz (2003). ‘Neural Substrates of
Tactile Imagery: A Functional MRI Study’. Neuroreport, 14 (4): 581–5.
Zacks, J. M., E. Levy, B. Tversky and D. J. Schiano (2002). ‘Graphs in Print’. In M.
Anderson, B. Meyer and P. Olivier (eds), Diagrammatic Representation and
Reasoning, 187–206. London: Springer.
Zagzebski, L. (2019). ‘Toward a Theory of Understanding’. In S. R. Grimm (ed.),
Varieties of Understanding: New Perspectives from Philosophy, Psychology, and
Theology, 123–36. New York: Oxford University Press.
Zatorre, R. J., A. R. Halpern and M. Bouffard (2010). ‘Mental Reversal of Imagined
Melodies: A Role for the Posterior Parietal Cortex’. Journal of Cognitive Neuroscience,
22 (4): 775–89.
Zeimbekis, J. (2010). ‘Pictures and Singular Thought’. The Journal of Aesthetics and Art
Criticism, 68 (1): 11–21.
Zeimbekis, J. (2015). ‘Seeing, Visualizing, and Believing: Pictures and Cognitive
Penetration’. In J. Zeimbekis and A. Raftopoulos (eds), The Cognitive Penetrability of
Perception: New Philosophical Perspectives, 298–327. Oxford: Oxford University Press.
Zeman, A. Z. J., S. Della Sala, L. A. Torrens, V.-E. Gountouna, D. J. McGonigle and R.
H. Logie (2010). ‘Loss of Imagery Phenomenology with Intact Visuo-Spatial Task
Performance: A Case of “Blind Imagination”’. Neuropsychologia, 48: 145–55.
Zeman, A., M. Dewar and S. Della Sala (2015). ‘Lives without Imagery – Congenital
Aphantasia’. Cortex, 73: 378–80.
Zeman, A., F. Milton, S. Della Sala, M. Dewar, T. Frayling, J. Gaddum, A. Hattersley, B.
Heuerman-Williamson, K. Jones, M. MacKisack and C. Winlove (2020). ‘Phantasia
– The Psychological Significance of Lifelong Visual Imagery Vividness Extremes’.
Cortex, 130: 426–40.
Zhang, J. (1997). ‘The Nature of External Representations in Problem Solving’. Cognitive
Science, 21: 179–217.
Zhao, F., W. Schnotz, I. Wagner and R. Gaschler (2020). ‘Texts and Pictures Serve
Different Functions in Conjoint Mental Model Construction and Adaptation’.
Memory and Cognition, 48 (1): 69–82.
 Index

Aasen, S. 105 Bildobjekt 131, 193 n.2

Abel, G. 19 binding problem 42
Abell, C. 32, 96, 126, 185 n.4 bistable figures 110, 174, 178
Abid, G. 179 Black, M. 191 n.11
Abrahamsen, A. 17 black hole picture 16, 75, 101–3, 105–10,
algebra 77, 190 n.4 119–20, 122, 127–8, 132–3, 151–2,
algorithm 77, 79, 85–6 154–6, 165, 182
Ambrosio, C. 9 Blackwell, S. E. 10
Amit E. 8 Block, N. 19, 175
amodal completion 178–9, 195 n.8 Blumson, B. 32, 49, 96, 144
Anderson, J. R. 195 n.5 Bocanegra, B. R. 6
Antonietti, A. 191 n.10 Boehm, G. 3
Aphantasia 177–8, 185 n.3 Bordwell, D. 22
Armstrong, D. 107, 131, 186 n.2, 191 n.2 Botterill, G. 17
Arnheim, R. 187 n.1 Boumans, M. 138
associationism (associationist theory of Bourne, L. E. 186 n.1
thought) 46–50, 187 n.8 Bradburn, N. M. 137
Ayers, M. 188 n.2 Braddon-Mitchell, D. 45
Braine, M. D. S. 12
Baddeley, A. D. 56 Briscoe, R. 11
Baillargeon, R. 189 n.3 Brown, J. R. 7
Ball, D. N. 137 Buccella, A. 114
Ballung concepts 138 Budd, M. 11, 105, 144
Bantinaki, K. 32, 126 Bueno, O. 95–6
Barsalou, L. W. 19, 57–8, 60 Burge, T. 9, 19, 41, 58, 103, 164–5, 194 n.3
Bartolomeo, P. 195 n.6 Burke, L. 108
Barwise, J. 7, 22 Byrne, R. M. J. 6
Bauer, M. I. 21
Bayne, T. 27 calibration 141–2
Beauchamp, M. S. 189 n.4 Camp, E. 25, 45, 49, 165, 171, 187 n.6
Bechtel, W. 17, 19, 185 n.4 Candlish, S. 19
Beck, J. 169 canonical decomposition 43–5, 164–5,
Bedny, M. 189 n.6 194 n.3
behaviourism 18, 47, 50–1 Cantlon, J. F. 189 n.6
Beilock, S. L. 22 Carey, S. 19, 61, 189 n.3
Belting, H. 3 Carruthers, P. 17
Bennett, M. 18, 64–5, 69 Cartwright, N. 137, 150
Bensafi, M. 10 Carvalho, J. M. 5
Ben-Yami, H. 13 Casati, R. 21, 106, 164
Berman, D. 7 Cavedon-Taylor, D. 175
Bermúdez, J. L. 12–13, 110 Chambers, D. 174
Bernecker, S. 192 n.4 Chang, H. 135, 137, 141
226 Index

Chappell, V. 188 n.2 indeterminacy of 30–2, 42, 158, 160,

Chevalier, J.-M. 9 163
Church, J. 185 n.3 non-propositional 13, 15, 20, 25–6,
Church-Turing thesis 85 48–9, 186 n.5
circumstances of evaluation, see context, particular 62, 89, 130, 192 n.1, 192
of evaluation n.3, 193 n.5
Clark, A. 59 phenomenological 11, 156
Clement, C. A. 6 propositional content 20, 25, 30,
Collins, C. 10 37–9, 66, 163, 169
compositionality 43–6, 48, 53, 63, 72, context
149, 164, 167, 171 communicative 31–3
concept formation 19 context (circumstances) of
connectionism 47 evaluation 124–5, 128–9, 160, 162
construction intensional 30, 48
abortive 85–6, 89, 146–7 modal 30, 159
concept of 14, 16, 66, 78–84, 90–1, propositional context 37
98, 101, 113, 121, 143, 146, 164, convention 9–10, 103, 118, 136, see also
182, 190 nn.2–5 iconic convention
construction invariants 16, 101, 111, Cooper, L. A. 150, 154
113–22, 129–30, 134, 153, 155, Coopmans, C. 21
166–7, 177, 180, 182, 192 n.6 coordination problem 141
definition of 83–9 correctness
indeterminable 146–8, 194 nn.10–11 accuracy conditions 29, 123, 133–4,
indetermined 194 n.10 136, 162
operation of 73, 75, 83–6, 90, 97–9, correctness conditions
114, 119, 123, 130, 146, 161, 165–6, (standards) 24, 70, 79, 87, 124,
177, 183 133–6, 148, 151, 157
parameters of 83–4, 86–8, 91, precision conditions 134, 136, 148
113–15, 117, 119, 130–1, 141, 146, robustness conditions 135
157, 165, 167, 180 trueness 134–6, 148
properties of 89–91, 95, 98–9, 120, Crane, T. 13, 29, 35, 37, 143
129, 131, 147, 159, 161, 192 n.1 criterion-less identification 160, 163
results of 83–4, 90, 141, 190 nn.5–6 Croijmans, I. 10
rules of 14, 16, 75, 79–82, 86–7, Cromwell, P. 76
89–91, 93–9, 113, 116–17, 120–2, cross-ratio 112–14, 120, 133
129–30, 139–40, 147, 149, 153–5, Cummins, R. 123, 126, 192 n.1
161, 163–6, 172, 176–80, 182–3, Currie, G. 185 n.4
187 n.7, 190 n.6
units of 83, 88, 130 Davidson, D. 36, 43, 131, 159
content Dawes, A. J. 185 n.3
bare bones 33–4 Defeyter, M. A. 8
diagram’s 78–9, 86–8, 91, 95, 99, 113, Dehaene, S. 189 n.6
182, 190 n.6 DeLoache, J. 186 n.4
distinctively sensory 11, 156 demonstrative files 108–9, see also
fleshed-out 33–4 object-file
general 130, 192 n.1, 192 n.3 demonstratives 16, 101, 106, 108–10,
iconic 18, 20, 25, 29–30, 37, 58, 86, 122, 182
129–34, 142, 150, 158, 160–4, 166, Denis, M. 25
169, 172, 176, 180, 182, 193 n.1, Dennett, D. C. 13–15, 19, 102–3, 128,
193 n.5 137, 167, 186 n.2
 Index 227

depiction Ernst, B. 144

experiential theories of 11 Ertz, T.-P. 79
intention-based accounts of 33, Etchemendy, J. 7
126–7 Evans, G. 40–1, 43, 45, 101, 106, 108,
make-believe theory of 12 110, 184, 186 n.6, 193 n.4
recognition-based theories of 104–5, Evans, J. St B. T. 186 n.1
121 exemplification 78, 80, 125, 129, 136,
resemblance-based theories of 75, 176–7
95–6, 98, 105–6, 124, 182 (see also
resemblance) Falmagne, R. J. 6
De Regt, H. W. 17, 151–2 Fazekas, P. 189 n.5
descriptionalism 174 Ferguson, E. S. 5, 17
description-based semantics 16, 75, see Finke, R. A. 11, 174
also identification by description Finkelstein, L. 137
De Toffoli, S. 79 Firestone, C. 59, 175
Devitt, M. 13, 44, 159 Fish, W. 191 n.2
diagrammatic reasoning 9, 21, 91, Fodor, J. 3, 25, 31, 43–4, 46, 48, 50, 112,
186 n.2 164, 166, 168, 194 n.3
diagrams Francis, G. K. 145
approximate definition of the content Frege, G. 13–14, 18, 34–5, 46, 48, 50, 66,
of 86–7, 129 183, 185 n.5, 190 n.3
token-diagram 89, 98, 130, 191 n.8 Frege-Davidson’s Argument 30, 34–6,
type-diagram 89, 130, 191 n.8 39, 42–3, 98, 149, 181
dialetheism 146 French, S. 95–6
Dilworth, J. 24 French, S. R. D 116
discursive representations 15, 25, 44, Frigg, R. 6, 129, 191 n.13
56, 73 Frixione, M. 32
Djordjevic, J. 10
Dokic, J. 107 Galton, F. 56
Dominowski, R. L. 186 n.1 Gardner, H. 17
Dorsch, F. 28 Gaskin, R. 42
drawings 2, 5, 12–13, 33, 38, 68, 87, 89, Gauker, C. 19, 25, 57, 60, 62–3, 108,
94, 109, 123, 126, 150, 161, 167, 175 150–1, 192 n.3
Dresner, E. 13, 137 Gendler, T. S. 17
Dretske, F. 12, 29, 168 generality constraint 43–6
dual-process theories 47 General Problem Solver (GPS) 46
Dummett, M. 14, 35 Gentner, D. 35
Gersel, J. P. 41
Earnshaw, R. A. 5 gestalt psychology 50–1
Eaton, M. 24, 29 Giaquinto, M. 7, 17, 91, 191 n.10
Eco, U. 44 Giardino, V. 7, 9, 22, 79, 103
Egan, D. 32 Gibson, J. J. 111–12, 114
Elgin, C. Z. 4, 102, 113, 126, 150, 152, Giere, R. N. 135, 191 n.13
175–6, 193 n.2 Gilbert, A. N. 10
Ellis, R. D. 185 n.2 Giovannelli, A. 107
Elpidorou, A. 143 Goldberg, B. 31
epiphenomenal nature of images 17–18, Goldin-Meadow, S. 6, 22
20 Goldstone, R. L. 58
epistemological challenge 28–30, 63, Gombrich, E. H. 11, 17, 98, 112
149, 151 Gooding, D. C. 5
228 Index

Goodman, N. 4, 12, 19, 32, 65, 70, 99, literary (metaphor) 10, 12
103, 124–6, 129, 168, 175–6, 184, measurement-theoretic account
191 n.1, 193 n.4 of 137–42
Goodwin, W. 29 olfactory 10, 12, 139–40
Grandin, T. 21 tactile 10
Green, E. J. 44, 59, 108, 114, 121 imageless thought controversy 18
Greenberg, D. L. 17 image-propositions 32
Greenberg, G. 9, 29, 103, 123–4, 192 n.1 imagistic cognition 62–3
Gregory, D. 11, 151, 156, 186 n.5 imagistic theory of thought 13, 19, 21–3,
Gregory, R. L. 144 25–6, 55, 57–9, 73, 149, 158, 175,
Grzankowski, A. 13, 37 183, 189 n.3
Guillot, A. 10 impossible fork 124, 143, 145–7
Gurr, C. 22 indexicals 132, 136, 152, 154, 165
indices 9–10, 108, 138
Hacker, P. 18, 64–5, 69 informative richness 169
Halpern, A. R. 10 Ingarden, R. 193 n.2
Harshman, R. 188 Inkpin, A. 112
Haugeland, J. 12, 33, 168 instrumental nature of images 17–18,
Hawthorne, J. 41 20, 22, 26, 28, 181
Heck, R. G. 29, 35, 43 intentionality 39–40
Hegarty, M. 21, 79 Intons-Peterson, M. J. 7
Heuer, F. 174 irreducibility thesis 15, 19–20, 23, 25–6
Hintikka, J. 37 Irvine, E. 185 n.2
Hinton, G. 174 Isaac, A. M. C. 117
Holyoak, K. J. 186 n.1, 188 n.1 Ishai, A. 174
homomorphism 191 n.12, 193 n.7 isomorphism 128, 191 n.12
Hookway, C. 9 Ison, M. J. 116
Hope, V. 106–7 Ittelson, W. H. 174
Hopkins, R. 11, 105, 127 Izard, V. 189 n.6
Hubbard, T. L. 8, 10
Hume, D. 47, 58, 187 n.8, 188 n.2 Jackson, F. 45
Hummel, J. E. 107, 192 n.5 Jäkel, F. 51
Husserl, E. 131, 193 n.2 Jakubowski, K. 10
Hyman, J. 80, 99, 105, 124, 193 n.4 James, M. 111
Johnson, K. 194 n.2
iconic convention 120–2, 124–5, Johnson, M. 185 n.2
129–33, 142, 194 n.8 Johnson-Laird, P. N. 6, 22, 188 n.1
iconic denotation 126–9, 131–2
iconic difference 3 Kamermans, K. L. 174
identification by description 106, 110 Kan, I. P. 58
image (icon) Kant, I. 27, 66, 70, 184, 187 n.11, 190 n.4
auditory 5, 10, 12–13, 139–40, 165 Kaplan, D. 33, 37, 40, 110, 128, 155
emotional 10 Kasem, A. 188 n.2
external (pictures) 9, 12, 185 n.4 Kaufmann, G. 19
gustatory 10, 164, 170 Kennedy, J. M. 144
haptic 10 Keogh, R. 177, 185 n.3
iconic representation 3, 8–12, 21, 42, Khalifa, K. 152
44, 99, 121, 164, 168–73, 178, 180, Kiefer, M. 58
183, 192 n.3 Kind, A. 28
impossible 123–4, 142–8, 194 n.11 Kinzler, K. D. 189 n.3
 Index 229

Kirsh, D. 21, 185 n.4 Lombardii, A. 32

Kitcher, P. 21, 29, 169 Lopes, D. M. 4, 10, 104–5, 111–12, 127
Kjørup, S. 24 Lowe, D. G. 116
Klatzky, R. L. 10 Lowe, E. J. 185 n.2, 188 n.2
Knauff, M. 6 Lupyan, G. 35
knots
knot diagrams 16, 75–9, 86, 120, 182, Macbeth, D. 7
189 n.1 McCarthy, R. A. 109
knot invariants 77 McDowell, J. 191 n.2
knot theory 75–9 McGinn, C. 168, 174
knowledge Machery, E. 58, 189 n.6
discriminating 40 MacKisack, M. 173
imagistic 24, 149–58, 180, 182 McNeill, D. 6
knowledge-by-acquaintance 151 Maglio, P. 185 n.4
non-propositional 80, 132, 150–1 Mahon, B. Z. 189 n.6
phenomenal knowledge 151 Maley, C. J. 168
propositional knowledge 24, 28–9, Malinas, G. 36–7
79, 86, 110, 150–1, 155, 183 Mancosu, P. 7
Kobayashi, M. 10 Mandelbaum, E. 49
Kosslyn S. M. 17, 56, 173–5 Manders, K. 7
Kosso, P. 152 Manley, D. 41
Kozak P. 22, 194 n.4 Marcus, R. B. 13, 137
Kozhevnikov, M. 21, 188 n.1 Margolis, E. 43
Krantz, D. 193 n.7 Martin, A. 58, 189 n.4
Kripke, S. A. 31, 126 Mast, F. W. 174
Kulpa, Z. 7, 143 mathematical proof 91, 190 n.2, 191
Kulvicki, J. 3–4, 10, 12, 24–5, 33, 37, 93, n.10
116, 144, 168, 171, 186 n.3 Matthen, M. 37, 45, 110
Kvanvig, J. L. 152 Matthews, R. J. 13, 137
Maynard, P. 137, 185 n.4
Lakoff, G. 185 n.2 measurement
Langacker, R. W. 185 n.2 concept of 127, 131, 134, 137–42,
Langland-Hassan, P. 24 156
Larkin, J. H. 21–3, 28, 169 devices 14–15, 26, 75, 127–8, 137,
Latour, B. 21 142, 148–50, 152–3, 156–8, 170,
Laurence, S. 43 176, 182–3, 193 n.8
Lehrer, K. 80 indications 156
Lemon, O. 22 invariants 118
Lewis, D. 168, 189 n.8 outcomes 135, 138–40, 156
Lieberman, K. 56 predicate 158–9
Lindsay, R. K. 22 representational theory of
Lipton, P. 151 measurement (RTM) 193 n.7
List, C. J. 32, 187 n.1 mental imagery
Liu, Z. 110 concept of 11, 19, 173–80, 184
Locke, J. 50–1, 57–60, 66, 73, 175, 188 dilemma of 175
nn.2–3 mental imagery debate 3, 7, 173–4,
Loewenstein, J. 35 186 n.2
logical form 5, 30, 34–7, 39, 42, 136, photographic model of 175
149–50, 158, 166, 169, 172, 181 weak perception theory of 174–5,
logic of showing 3 178
230 Index

mental images neo-empiricist consensus 58

concept of 12, 173–80, 185 nn.3–4 neo-Lockean approach 15, 56–7, 60, 63,
constructive account of 176, 179 68, 175
experiential account of 176 Nersessian, N. 5
unconscious 11, 176–7 Newall, M. 11, 105
mental representation format Newell, A. 46, 50
analog 168–70, 182 Newton, N. 10
concept of 168 Nguyen, J. 129
digital 168–70, 183 notation
domain-generality of 170, 172–3, concept of 22
183 Conway notation 77
domain-specificity of 170, 172–3, Dowker notation 77
178, 180, 183 Novitz, D. 3, 105
hierarchical information architecture Nunberg, G. 38
of 171–3, 183 Nyíri, J. C. 31, 185 n.2
holistic nature of 171–3, 180, 183
non-hierarchical information Oberlander, J. 191 n.10
architecture of 171–3, 183 object-file 108–9
mental rotation tasks 21, 155, 167, 175, O’Brien, D. P. 12
177, 179 operational approach 15, 56, 67–70,
Merricks, T. 13 74–5, 181, 189 n.7
Mersch, D. 3, 36 optical information 111–12, 120
metaphysical challenge 43–6, 53, 63, Otis, L. 188 n.1
164, 167 Overgaard, S. 106
Metzler, J. 17, 56, 154, 173
Meynell, L. 17 Pagin, P. 44
Michaelian K. 192 n.3 Paivio, A. 17, 56, 173, 188 n.1
Mitchell, W. J. T. 3 Peacocke, C. 11–12, 43–4, 107–8, 168–9
Miyake, A. 56 Pearson, J. 56, 174, 177, 185 n.3, 195 n.6
Molitor, S. 8 Pecher, D. 58, 189 n.4
Montague, M. 13 Peirce, C. S. 4, 9–10, 91–2, 184, 187 n.11,
Morgan, M. S. 138 190 n.4, 191 n.8
Morrison, M. 138 perception, see also thought, thoughts-
Morrison, R. G. 186 n.1 perception border
Mortensen, C. 142, 145–6 cognitive impenetrability of 175
Mößner, N. 5, 25, 63, 129, 151 object perception 106–7
Mullarkey, J. 19 perceptual illusions 144
Müller, A. 3 perceptual invariants 111, 114
Mumma, J. 7 perceptual recognition 106–8, 111,
Murez, M. 108 116, 179, 191 n.2
that-perception 106–8, 112
Nail, T. 10 Perini, L. 29
Nanay, B. 4, 10, 174, 178, 185 n.3, 189 Pero, F. 191 n.12
n.5, 195 n.8 Perry, J. 152
Narens, L. 137 Peterson, M. A. 174
nativism 57, 189 n.3 pictorial fallacy 36–9, 90, 159, 186 n.4
natural generativity 104, 121, 166 pictorialism 173
Neander, K. 105 picture principle 44, 164
Necker cube 91, 143, 163 picture superiority effect 8
Nelsen, R. B. 7 Pietarinen, A.-V. 9, 186 n.2, 191 n.12
 Index 231

Podro, M. 11 referent 16, 37, 41, 89, 123–5, 129–36,

predication (predicative function) 39, 139, 142, 148–9, 160–2, 164–7, 182,
41–2, 48, 158–62, 170–1, 186 n.4, 192 n.1
193 n.4 Reidemeister moves 76–9, 85, 87
Price, H. H. 4, 19, 69, 187 n.1 Reidemeister’s theorem 76
Priest, G. 143–6 Reisberg, D. 7, 174
principle of charity 131 Rescorla, M. 170, 187 n.9
Prinz, J. J. 57–8, 60–2 resemblance 9, 32–3, 36, 92–9, 105, 122,
private language argument 31, 65 124, 142, 157–8, 161, 182, 191 n.11,
Proclus 81, 190 n.2 see also depiction, resemblance-
projective geometry 82, 112–14, 116, based theories of
120 Reutersvaard-Penrose triangle 143–6,
properties 194 n.9
first-order 92, 94, 96, 117 Rey, G. 13
second-order 93–5, 105, 117 Rheingold, H. R. 5
propositional attitude 1, 12–13, 29, 46, Richardson, J. T. E. 187
51, 107 Richtmeyer, U. 31
propositionalism (propositional theory of riddle of style 98, 191 n.13
thought) 46, 48–9 Riemann, B. 82, 114
proxytypes 60–1, 63 Rocke, A. J. 188 n.1
Pulvermüller, F. 58 Rogers, Y. 8, 174
Putnam, H. 93, 124 Rollins, M. 19
Pylyshyn, Z. W. 3, 7–8, 18, 25, 43, 45, 48, Roser, A. 32
55, 59, 108, 112, 154, 166, 168, 173 round square 147
Rozemond, M. 41
Quilty-Dunn, J. 44, 49, 108, 171 rules, see also construction, rules of
Quine, W. V. O. 38, 40, 69, 194 n.2 rules of interpretation 80–1, 87
Quiroga, R. Q. 116 rules of production 80–1, 87
Runhardt, L. 137
Raftopoulos, A. 59, 175 Russell, B. 19, 32, 40, 55, 187 n.1
Ragni, M. 6 Russell’s Principle 40–2, 108
rationality 36, 46, 48–9, 65, 150 Ryle, G. 5, 18, 27, 64–6, 69, 71, 179,
Rayo, A. 83 187 n.10
Recanati, F. 108
Received View 15, 27, 46–9, 52–3, 181, Sainsbury, R. M. 35
186 n.1, 187 n.8 Salis, F. 6
recognition (recognition-based Salje, L. 194 n.2
identification) 16, 75, 80, 101–24, sample 23, 42, 125–6, 135–6, 139, 157,
129–31, 152, 154, 156, 166, 179–80, 172
182–3, 191 n.2, 192 nn.4–5 Sartre, J.-P. 174
reference Sartwell, C. 104
causal models of 126 Scaife, M. 8, 174
deferred 38 Scarry, E. 10
direct 9, 12, 105, 108, 111, 116, 122, Schier, F. 11, 99, 104–5, 124
130–2, 163 Scholl, B. J. 59, 175
iconic 124–6, 134, 136, 148, 182 Schooler, J. W. 50
non-rigid 133, 160 Schöttler, T. 106
rigid 133, 160 Sebeok, T. A. 10
two-dimensional model of iconic 16, seeing-in 96, 124, 131, 192 n.6
75, 123–33, 147, 182 Sellars, W. 1, 3, 13–14, 19
232 Index

semantical challenge 39–43, 149, 158, systematicity 15–16, 31, 43–6, 48, 53, 63,
186 n.5 109, 112, 137–8, 149, 164–7, 180–2,
Shabel, L. 190 n.4 186 n.1, 186 n.6
Shah, P. 56
Shaver, P. 6 Tal, E. 135, 141, 156
Shepard, R. N. 17, 19, 56, 150, 154, 173, target 16, 83, 87, 89, 108–11, 114–16,
188 n.1 123–33, 135–6, 141–2, 146–7, 149,
Sheredos, B. 5, 185 n.4 156, 160–3, 182, 191 n.13, 192 n.1,
Shimojima, A. 22, 186 n.2 193 n.2, 193 nn.4–5, 194 n.11
Short, T. S. 9 Tarski, A. 31
Siewert, C. 107 Teller, P. 135
Simon, H. A. 21–3, 28, 46, 50, 169 Terrone, E. 111, 127, 129
Skinner, B. F. 51 Teufel, C. 189 n.5
Slezak, P. 2, 64, 174–5 Thagard, P. 10, 17, 21, 188 n.1
Sloman, A. 5, 28 Thielen, J. 178
Sober, E. 44 thinking
Soles, D. 188 n.2 animal 19
Solomon, K. O. 43 imagistic 2, 4, 11–15, 18, 20, 23–8,
Solt, K. 29 34, 36, 42–3, 48–9, 51–3, 55–60,
Sorensen, R. 142 62–3, 68, 70, 73, 75, 149–81, 187
space n.1, 188 n.1
informational 83, 85–6, 90, 99, 113, individual differences in 188 n.1
117, 121, 129 language-like 2, 5, 12–13, 18, 44, 46,
logical 14, 51–2, 83, 86, 105, 114, 62, 173, 181
127, 129, 131–2, 152–4, 157, 165, measurement-theoretic account of
176, 180, 182, 191 n.7 13–15, 149, 179–80, 183
perceptual similarity 62–3, 108 metaphors of 49–52
physical 14, 51–2, 83, 86, 105, 127, propositional 12–13, 25, 182
131–3, 152–4, 157, 165, 180, 182, thinking styles 188 n.1
191 n.7 Thomas, N. J. T. 11, 185 n.2
spatial representations 186 n.2, 194 n.11 thought
Spelke, E. S. 189 n.3 bearer of 15, 20, 26–8, 31–2, 45, 53,
Squires, R. 104 56, 65, 67, 69–70, 164
standard metre, 118, 133, 135, 140, 162, imagistic 2, 4, 11, 15, 18, 20, 23–7,
178, see also measurement 42, 51, 56–60, 63, 68, 70, 73, 75,
Stanley, J. 151 149, 181, 187 n.1
Stegmüller, W. 66 thought acquisition 47, 57
Stenning, K. 12, 22, 191 n.10 thoughts-perception border 59–64,
Sternberg, R. J. 6 73, 179
Stich, S. P. 35 Tichy, P. 82, 190 n.3
Stjernfelt, F. 9, 22, 186 n.2 Traditional View 15, 39, 43, 73, 75, 140,
Strawson, P. F. 109 142, 148–50, 157–8, 161, 164, 175,
structural similarity 93–6, 98, 116 181–2
Suárez, M. 32, 95, 191 n.12 translatability thesis 15, 23–6, 49
Swoyer, C. 17, 93, 131, 135, 137, Troscianko, E. T. 10
186 n.2 truth
symbols (symbolic representations) 6, truth-bearers 28–9, 31, 34–5, 150
9–10, 18, 22, 25, 31, 69, 72, truth-conditions 28, 31, 36, 38, 135,
90, 104–5, 121, 129, 187 n.7, 185 n.5
190 n.4 truth-makers 29
 Index 233

T-schema 135, 172 White, A. R. 188 n.2

Tufte, E. R. 5 Wiesing, L. 193 n.2
tunnel effect 108, 119 Willats, J. 103
Tversky, B. 21, 165, 187 n.7 Williamson, T. 28, 150–1, 157
Twardowski, K. 66, 171 Winn, W. 8
Tye, M. 29, 151 Wittgenstein, L. 18, 30–4, 45, 64–6,
162–3, 184
understanding 151–8, 165–6 Wittgenstein-Ryle’s sceptical
argument 15, 56, 64–7, 73–4, 90
Vallée-Tourangeau. F. 6 Wittgenstein’s argument from content
van Fraassen, B. C. 4, 96, 117, 132, 138, indeterminacy 30–4, 39, 42, 45, 96,
141, 153 98, 149–50, 161, 163, 181, 187 n.1
Varzi, A. 21, 29, 164, 169 Wollheim, R. 11, 32, 44, 96, 104–5, 119,
Vernazzani, A. 191 n.2 124, 131, 187 n.1
Viera, G. 10
visual impedance effect 6 Yolton, J. W. 188 n.2
Voltolini, A. 11, 144 Yoo, S.-S. 10
von Neumann, J. 170 Yule, P. 12

Walton, K. L. 12, 99, 105, 124 Zacks, J. M. 185 n.1

Ware, C. 22 Zagzebski, L. 151
Warrington, E. K. 109, 111 Zatorre, R. J. 10
Watson, D. 5 Zeimbekis, J. 29, 59, 110
Watson, J. B. 18 Zeman, A. Z. J. 11, 177, 185 n.3
Westerhoff, J. 36 Zhang, J. 21
Westerståhl, D. 44 Zhao, F. 2
234
 235
236
 237
238