OUR KNOWLEDGE OF COLOUR                            

 

  Mohan Matthen

University of British Columbia

 

Scientists are often bemused by the efforts of philosophers essaying a theory of colour: colour science sports a huge array of facts and theories, and it is unclear to its practitioners what philosophy can or is trying to contribute.  Equally, philosophers tend to be puzzled about how they can fit colour science into their investigations without compromising their own disciplinary identity: philosophy is supposed to be an a priori investigation; philosophers do not work in psychophysics labs – not in their professional capacity, anyway. 

These inter-disciplinary barriers arise out of misunderstanding.  Philosophers should not so much attempt to contribute to empirical theories of colour, as to formulate philosophical theories of colour.  Philosophy is concerned with appearance and reality, object and property, function and representation, and other such fundamental categories of ontology and epistemology.  Philosophical theories attempt to fit colour into these categories; such theories do not compete with colour science.  However, fitting colour into philosophical theories means dealing with colour as it really is – and one cannot know what it is without consulting the psychologists.  That is why philosophers need an up-to-date understanding of psychological theories of colour.  Equally, psychologists, who typically show a keen interest in questions concerning the reality and knowability of colour, and who are capable of devising clever experiments to discover whether colour fits this or that specification of reality, knowability, etc., need to base their opinions on sound philosophical practice.  Their opinions too are worth very little when they misuse fundamental philosophical categories. 

In this paper, I am concerned with how the ontology and epistemology of colour relate to the psychology of colour-perception.  The empirical facts to which I appeal are not the subject of controversy: they are almost all reported in the path-breaking papers on the opponent-process theory that Dorothea Jameson and Leo Hurwich published in the nineteen-fifties;[1] most of these facts have been easily available to philosophers at least since Larry Hardin’s book (1988) on colour.  The philosophical interpretation of the facts remains, however, a matter of sharp and unresolved controversy.  In section I, I’ll briefly summarize the controversy, and relate how it has been deepened by the articulation of a new epistemological constraint by Paul Boghossian, David Velleman, and Mark Johnston around 1990.  Then I will turn to the task of giving an account of colour ontology that both satisfies a reasonable version of this constraint, and bridges some of the conflicting intuitions concerning the empirical facts about colour-perception. 

 

      I.  A PROBLEM CONCERNING THE ONTOLOGY OF COLOUR

 

A.  Conflicting Intuitions Concerning Colour Ontology

The colour appearance of a thing is highly variable with illumination, contrast, adaptation, and individual physiological constitution, much more so than philosophers have traditionally supposed.  For example, colours like brown, which have black or white as a component, are seen only when there is a contrast available.  Thus, an object that looks brown when you look at it in the context of other objects might look orange when you isolate it by looking at it through a reduction tube.  Now, what colour is such an object, really?  Is it brown because it looks that way in a normally multi-coloured scene?  Or is it orange because it looks that way when we remove the highly variable influences of contrast?  There seems to be no principled way to decide in favour of one of the two appearances.  Both viewing conditions have some claim to metaphysical privilege.  Moreover, both are within the range of conditions that could be considered “normal” in evolutionary terms.  Thus, the variation in appearance seems nothing more than that.  The best one can say is that the object is brown when contrast is provided, orange when contrast is removed, and neither colour independently of viewing conditions.  On this account, colour is “subjective”, or “observer-relative”, in the sense that it is the product of an interaction between a perceiver and the distal environment, and cannot be attributed to that environment in isolation. 

On the other hand, colour appearances are much more constant than one might expect given the variability of the light array that reaches the human eye.  For example, a human face looks normally coloured even under heavy leaf cover, even though the light it reflects in such circumstances shows distinctly green in colour photographs.  Colour vision seems, then, to “correct” the look that things have in non-standard circumstances.  This kind of constancy suggests to many philosophers that colour vision evolved to detect a real property of things, a property they have independently of any appearance.  It seems, in other words, that human faces are not green no matter what signal they might be sending in “non-standard” circumstances, and that the colour they appear in a variety of more or less “standard” circumstances is the colour they really are.[2]  This is an “objective” account of colour.  It implies that we cannot just leave the variability of colour unresolved, as the subjectivist would urge.

Which should we take more seriously, the variability of colour appearance or its constancy?  Should we say that colour vision converges on certain invariant distal properties, and leave variation as an anomaly?  Alternatively, should we say that colours are appearance-relative, and hold that constancy is, though useful, an oddity – merely a by-product of certain properties of colour-sensitive cells in the retina, for instance? 

 

B.  An Epistemological Constraint on Theories of Colour

This dilemma is made all the more difficult by a recently articulated epistemological constraint that seems to challenge fundamental intuitions about appearance and reality, and sharply distinguishes colour from shape, motion, etc., the so-called primary qualities, and even (as I shall argue in section I C) from many other secondary qualities, for example, musical harmony.  Mark Johnston (1997, p. 138) phrases the constraint this way:

 

Revelation 

The intrinsic nature of canary yellow is fully revealed by a standard visual experience as of a canary yellow thing (and the same goes, mutatis mutandis, for the other colours).[3] 

 

What is it to know the “intrinsic nature” of a colour?  However we answer this question, Revelation seems to imply a quite powerful kind of realism.  The colours have to be independent of our sensations if they have “intrinsic natures”, or so it would seem.  At the same time, it implies that colours lack physical reality, for surely one cannot come to know the intrinsic nature of physical properties (like wavelength or reflectance) by bare experience – untutored vision has no knowledge of such things.[4]  But if the colours are not physical what are they?  Do they depend on our mental (or perceptual) reactions to things?  How does this differ from the subjectivist thesis? Revelation seems thus to deepen and inflame the conflict of intuitions noted in the last section. Nevertheless, it cannot be denied that there is an important truth that lurks somewhere in its vicinity.  We use colour vision to categorize things for identification, induction, signalling, and the like.  When we co-classify things by colour, we instinctively treat them as the same as far as these practices are concerned.  These practices are experience-based, in the sense that it requires no more knowledge to participate in them than to experience colours.  Presumably, animals use colour in comparable ways without the aid of linguistic concepts – a honeybee identifies flowers by their colour, an old-world monkey identifies fruit, and so on.  Colour categorization is a primitive ability we share with these animals.  Revelation is plausible if we construe it as asserting that this kind of categorization is unlearned and unconditioned, or at least learned only in highly predictable ways invariant across individual experience sets and across culture.  Colour-experience may or may not furnish us with knowledge of a property, then, but at the very least, it does give us the instinctive know-how needed for colour categorization. 

With this in mind, let us consider the following weaker version of the condition stated (and rejected) by Johnston:

 

Empiricist Codicil to Revelation

Experiences as of canary yellow things are

 

(a) sufficient for knowing the basis for co-classifying canary yellow things for the purposes of inductive inference (etc.), and

 

(b) sufficient together with experiences as of lime green things for knowing the basis for differentiating canary yellow things from lime green things. 

 

(And the same goes, mutatis mutandis, for the other colours.)[5]

 

In this weakened form, I take Revelation to be true.  (In the rest of this paper, I shall be referring to this weakened form of the thesis under the name of Revelation, except when I intend a contrast with the stronger principle.) 

Imagine, then, that you have never experienced canary yellow before.  A colour scientist arranges for you to have this experience by artfully arranging for you to have an after-image of that shade, or, even more diabolically, by electronically stimulating your brain in such a way as to produce in you a sudden flash of canary yellow, floating free of any object or shape.  You do not form any belief about the external world because of this self-evidently ephemeral experience.  You do not attribute the colour to any external thing.  Properly understood, the Empiricist Codicil implies that even this experience is somehow sufficient to reveal the basis for classifying things by a certain colour category.

The puzzling question is this: what kind of classification is this that its entire basis is fully revealed by a mendacious experience?  Moreover, how can such a classification be useful for externally validated practices such as object-identification and induction?  This sharpens the conflict of intuitions noted earlier.  If Revelation is taken to show that colour-experience fails to correspond to an independently existing thing, then what are we to make of colour-constancy?  And if we take colour-constancy to be evidence of the objectivity of colour, then how is it that colour, in contrast to so many other objective categories, is so easy to know, even by means of self-evidently non-veridical experiences?

 

C.  The Contrasting Case of Musical Harmony

From what we have said so far, one might think that if Revelation were correct for colour, it would apply to other “secondary qualities”.  And in a sense, this is right.  That is, it must be the case that if we participate in instinctive classificatory practices that involve these other secondary qualities, these practices too must be appearance-based.  Nevertheless, Revelation is more puzzling applied to colour than to musical harmony.  The reason may be surprising, and worth examining in a little detail. 

Before we get to that, however, we need a preliminary observation to avoid confusion.  With any sensory modality, the novice observer requires training in order to be sensitive to the character of her experience.  That is, it takes practice (and possibly some instruction) to recognize the presence of yellow in lime-green, the taste of raspberries in a wine, the component structure of a musical chord, and so on.  However, the need for this kind of “training” does not compromise Revelation.  True, one might not appreciate all that is present in an experience of canary yellow immediately upon having an experience as of it.  Still, that single experience may contain everything needed to understand the category of canary yellow (in the sense of the Empiricist Codicil).  By training the eye, the palette, or the ear, one does not come to have new, more complex, experiences.  Rather, one learns how to discriminate the components already present in the experiences one has had.  Or so I shall assume to keep Revelation in play. 

Now, just as is claimed for the case of colour, it may seem that the experience of musical harmonies must “reveal” something about their “intrinsic nature” (as Revelation would demand). Consider one’s auditory experience of a minor third.  After a period of ear training, one begins to recognize such things as the interval heard when the constituent tones of this chord are played in sequence, its relationship to other chords, the musical character of chord progressions in which it figures, and so on.  Now, it turns out that what one cognizes in this way has a counterpart in objective reality.  An ear-trained listener will find that her discriminations are confirmed by certain tests on a piano or other musical instrument; for instance, she will find that she hears a minor third just when one is played.  If not, she is in error.  There is something wrong with a listener who has an auditory experience as of a minor third when we sound C and E together.[6]  Her ear training (or just her ears) is shown to be deficient by her failing this objective test.  From these facts, we may conclude that

 

(1) there is a structure to one’s experiences of a chord that .  .  . 

 

(2) .  .  .  corresponds (perhaps imperfectly and incompletely) to objective structural characteristics of the chord itself.

 

The correspondence between the structure of one’s auditory experience and the sound one hears goes some distance, at least, towards explaining the counterpart of the Empiricist Codicil for the auditory grasp of chords. 

However, this conjunction of conditions does not hold of colour-experience.  The difference between the two cases does not lie in condition (1) above.  There is, as the experimental psychophysics of colour reveals, structure in our experiences of colour, for they are ordered in the dimensions black-white, red-green, blue-yellow, and brightness.[7]  The difference lies rather in (2): the experienced component structure of colour is largely a by-product of the way in which colour is processed by our visual systems, and does not correspond to the intrinsic character of coloured objects and lights. 

What I mean is this.  The physical counterparts of colour-experience do not share the component structure of colour-experience – for instance, there is no physical counterpart of green and of blue invariably present in physical instances of turquoise.  Indeed, turquoise, like most other colours, can exist in lights, in transparencies, and in reflective surfaces, and the principles of mixing are different in these diverse media.  One way to see this is to note that both turquoise lights and turquoise paints can be produced without mixing blue and green.  There is a mixture of minerals that is ground up and added to a vehicle to make turquoise paint.  There are salts that turn the flame of a Bunsen burner turquoise.  There is no physical mixing of blue and green in such instances of turquoise production.  Thus, the component structure of the turquoise experience tells us nothing about the constitution of physical turquoise.  Nor does it purport to: the experience of turquoise does not lead us to expect physical mixture.  This is how it differs from the musical example: a chord physically contains its component notes; listeners can be right or wrong about what they hear in this respect.  Thus, there is no transfer, as in the case of music, from the structure of reality to the structure of experience. 

We can now restate what is puzzling about Revelation even as weakened in the Empiricist Codicil.  Sensory categories, including colour, underwrite induction and other epistemic practices.  In order to be useful in these epistemic roles, one would think that they must correspond in some way to objective characteristics.  It seems to follow that colour-experiences must correspond, somehow, to objective colours.  But, on the face of it, they do not. 

 

         II.  TWO APPROACHES TO DEFINING THE COLOURS

 

A.  The Standard Viewing Conditions Approach

Let us now introduce the notion of a colour-look.  Recall first that colours can be arranged in a three dimensional similarity-space by systematically collating the similarity and discrimination judgements of observers presented with colour samples.[8]  Colour-looks can be identified with minimal regions in this similarity-space, that is, regions so small that the colours they contain are indiscriminable.  I shall speak of environmental objects presenting observers with colour-looks. 

 

x presents observer O with colour-look L at t if and only if x occupies region L in O’s colour similarity-space at t.[9]

 

Colour-looks are psychological in character: they are ways in which objects are subjectively experienced by observers, and derive from these observers’ abilities to match and discriminate objects with respect to colour.  Philosophers, especially empiricists, standardly define colour-attributions in terms of colour-looks.  (We shall see how to do this in a moment.)  At first sight, this helps with Revelation.  For, as just said, colour-looks are subjective.  It follows that we know them just by having colour-experiences.  So, if we define colours in terms of colour-looks, we gain some hope of piggybacking our knowledge of colour on our knowledge of colour-looks.

There is, however, an obvious difficulty in basing colour-attribution on colour-looks.  The look that a particular object presents to an observer varies with viewing conditions, while colour seems to be an enduring feature of objects.[10]  Philosophers tend to approach this problem by relativizing colour-looks to various viewing conditions.  Let us suppose that the following statement is fully specified in this respect:

 

x presents colour-look L to observer O when x is illuminated by light of spectral distribution S of luminance L, with surround colours C1-Cn, when O is adapted to light of spectral distribution S' and x is displaced from the centre of O’s retina by θE.

 

Fully specified colour-looks are invariable because all the relevant variables are fixed.  The thought is that we can therefore use fully specified colour-looks to specify object-colour.

This approach fails because the problem of Revelation is displaced to the question of how the conditions so specified can be known.  Consider the appeal to “standard conditions” to provide the needed specifications, as in the following principles of colour-attribution.

 

Dispositional Principle of Colour Attribution

“x is L” is true if and only if x has the disposition to create in a standard observer the colour-look L in standard conditions.[11]

 

Counterfactual Principle of Colour Attribution

“x is L” is true if and only if x would create in a standard observer the colour-look L in standard conditions.

 

These principles become fully specified by providing details of standard observers and standard conditions. [12]  Such a specification might go something like this.  A standard observer is one who has trichromatic vision.  This observer must be white-light adapted, and fixating x in the centre of her retina.  x must be presented to such a viewer in a surround of neutral grey, illuminated by white light to luminance 10 mL.  The colour of an object is thus defined as the disposition to create the colour-look in question in viewing conditions other than those that might be immediately present.  The problem with this is obvious: even if the viewer knows, by visual experience, what the colour-look mentioned in the above attribution-principles is, how is she to know what “standard conditions” are?[13] 

Think again of the shimmering canary yellow film floating in the air.  Revelation implies that this ephemeral experience is enough to reveal what it is like for something to be canary yellow.  However, this visual experience contains no information about standard viewers and conditions, no means of determining to whom and in what circumstances a thing must look like this if it is to be canary yellow.  Indeed, your experience of the film does not contain any allusion to viewing conditions at all: you just do not know how the film would look in “standard conditions” or what standard conditions are for it.  The advantage of appealing to colour-looks was thought to be that experience gives us direct knowledge of them.  This advantage is negated by the appeal to viewing conditions other than those present to the viewer. 

 

B The “Semantic” Account of Colour

Colour-looks seem to give us direct knowledge of colour-properties, but we are finding it difficult to understand how this can be so, given the standard viewing conditions approach to defining colour-properties.  At the risk of repetition, let us summarize the difficulty.  The colour-look that an object presents varies with conditions of viewing.  While colour-constancy is an indication that the system is capable of compensating for this variability to some extent, there is still a range of circumstances, all “normal” in evolutionary terms, in which the same object will present a different look.  Thus, despite constancy, colour-looks do not vary one-to-one with any object-property.  The “standard conditions” approach seeks to negate variability by focussing on the conditions under which a colour-looks reliably indicates what colour-property an object has.  However, since the colour-vision system does not automatically signal whether or when it is operating within this range of conditions, the perceiver requires empirical knowledge to grasp and apply this definition of colour-properties.  This is the problem.  If one wants to preserve the instinctive grasp of colour premised in Revelation, one needs a connection between colour-looks and colour properties that demands less empirical knowledge. 

Luckily, there is such a relation between colour-looks and colour-properties available to help us.  Suppose that somebody shows you a coloured object, and asks you what colour-property it visually appears to have.  You need no collateral information about viewing conditions in order to answer this question.  You need such information only in order to know what colour something is, not to know what colour it looks.[14]  The standard conditions approach attempts to define a colour-property in terms of reliable indications of when a thing actually possesses that property.  We lack the information needed to assess when looks are reliable.  However, it seems as if we have direct knowledge of what colour-property a thing looks as if it has, when it presents a given colour-look.  If this is so, then we should be able to reach into the “looks” context, as it were – more technically, we can quantify into it – in order to specify colour-properties.  It helps us to communicate the property in question, and the look in question, if we remember that in English, and every other natural language, the look and the property have the same name. 

Along these lines, then, consider the following schema:

 

Look Exportation

Canary yellow is the colour-property something visually appears to have when it presents the canary-yellow-look.  More generally: L is the property something has when it presents the look, L (where L is the colour-property with the same name as L).

 

In the above schema, the “looks” in “looks L” does not betoken the relationship of reliable indication.  That is, it does not say what would clearly be false: that canary yellow is the colour you automatically attribute to x when x presents you with the canary-yellow-look.  Look Exportation should not be understood, then, in terms of any tendency to believe that something possesses a colour. 

What then does Look Exportation say about the relationship between colour-looks and colour-properties?  The following analogy with linguistic communication might be helpful.  Suppose somebody says to you, “I make one hundred thousand dollars a year”.  It takes quite a lot of empirical knowledge to figure out whether this statement is a reliable indication of the speaker’s income.  And even if you possess such knowledge, determining the likelihood of the speaker’s truthfulness depends on divining her circumstances.  Is she negotiating a salary offer?  Is she negotiating a divorce?  However, it takes only semantic knowledge, and no knowledge of either background or present conditions to determine what the speaker is telling you.  The semantic link between utterance and meaning bypasses listeners’ dispositions to believe the speaker and bypasses the conditions of utterance.  The connection between colour-looks and visual appearance is like this.  We said earlier, while discussing Revelation, that when we undergo a colour-experience, we know instinctively what colour classifications the experience licenses.  Look Exportation is a consequence of this instinctive link. 

Look Exportation is a better starting-point for our inquiry into the nature of our grasp of colour-properties than the world-to-look correlations studied by psychophysics.  It gives us a quite direct way of defining what it is for something to be canary yellow.  A thing is not canary yellow merely because it presents a canary-yellow-look.  But when something does this, it invariably looks canary yellow.  The look identifies a property, C, and a thing is C if it actually possesses this property.  In other words, something is canary yellow if it really is the way such a thing looks.  This gives us the following schema:

 

Fundamental Principle of Colour Attribution

“x is Col” is true (where Col is a colour term) if and only if x really is the colour something visually appears to be when it presents the Col-look. 

 

The Fundamental Principle introduces a “Really is the way it looks” operator.  This operator takes colour-looks as arguments, and yields colour-attribution conditions as values.  Its logical force is different from the “Would look that way in standard conditions” approach considered earlier, which tells us how something would look in specified or standardized conditions. 

The Fundamental Principle is analogous with the disquotation principle enunciated by Alfred Tarksi (1944).  Tarski observed that the removal of the quotation marks from ‘snow is white’ gives us a way of asserting what would be the case if ‘snow is white’ were true.  This observation led him to his famous principle:

 

The sentence ‘snow is white’ is true if, and only if, snow is white  (ibid., p. 343).

 

In much the same way as the disquotation principle, the “Really is the way it looks” takes advantage of our instinctive grasp of the colour-classifications licensed by a colour-look to specify a particular colour-property.  Tarski entitles his conception “the semantic conception of truth”, explaining that “Semantics is a discipline which, speaking loosely, deals with certain relations between expressions of a language and the objects (or ‘states of affairs’) ‘referred to’ by those expressions.  As typical examples of semantic concepts we may mention the concepts of designation, satisfaction, definition .  .  .” (ibid., p. 345).  Following him, I shall say that colour-looks designate colour-properties, and entitle this conception the “semantic” account of colour. 

The semantic account points to a very different way of thinking about colour-attribution than that which is implicit in the “standard conditions” approach, which (as I said before) looks for conditions in which colour-looks are especially reliable, or in some other way privileged.  This difference is analogous to that between an epistemic theory of truth, like Epicurus’s,[15] which refers to the conditions under which a belief is reliable, and a semantic theory of truth, like Tarski’s, which refers to the conditions designated by a sentence.  The Fundamental Principle re-orients our attitude towards the look-property relation away from reliable indication. 

The “semantic conception of colour” implicit in the Fundamental Principle is insufficient by itself actually to provide us with a philosophical account of the look-property relation.  All it does is to take advantage of our instinctively implicit knowledge of colour-attribution.  In order more fully to understand the relationship between colour-looks and colour-properties, we need to undertake to further tasks.  We must

 

First: explicate the nature of the semantic character of the relationship between colour-looks and colour properties.  In other words, we must show in what way colour-looks are like symbols that designate colour-properties. 

 

and

 

Second: give a more explicit account of which property each colour-look designates: in other words, make our implicit knowledge of colour-attribution more explicit. 

 

Given the demands of Revelation, we need to do this in a way that excludes collaterally acquired knowledge.  I undertake these tasks in the remainder of this paper. 

 

       III.  HOW COLOUR-LOOKS FUNCTION AS SYMBOLS

 

A.  Property-Designation in Measuring Instruments

Fred Dretske offers us a metaphor that proves useful in explaining the semantic character of the connection between colour-looks and colour-properties.[16]  He likens sensory systems to measuring instruments.  Every measuring instrument is associated with a measurement function, M, which connects its measuring states – its pointer readings, for instance – to measurement values, properties of the object being measured.  Sensory systems seem to be similar: the colour-vision system has states that seem, as we saw in the last section, to designate colour-properties.  Revelation implies that we are instinctively able to use the measurement function associated with sensory systems.

How are we able to use the measurement function of measuring instruments?  How do we know (a) what state the instrument is in, and (b) what property corresponds to that state? 

 

(a) Identifying the measuring state Measuring instruments are designed in such a way as to enable us easily to identify their measuring-states: for instance, they might have needles that point to markings on their faces, or some other type of display.  Each position of the needle marks a different measuring state.  (Imagine a pressure gauge that has lost its needle.  Such a gauge too has measuring states, but these are now difficult, if not impossible, to read.) 

 

(b) Knowing the measurement value of each measuring state  In order to use a measuring instrument, we need access to a key that tells us how to specify the values of various measuring states.  Often, we can read a linguistic expression of the object property off a transparent notation expressed by the gauge-display.  For instance, we might find numerical expressions like ‘14’ written on a pressure gauge, with a notation ‘pounds per square inch’ written across the bottom.  When the needle points to ‘14’, the notation on the dial enables us to express the object property as ‘fourteen pounds per square inch’.  (A notation in newtons per square meter would express the same property but in different terms.)  Notations that are more opaque are also possible.  A graphic notation would be an example.  A square might indicate 14 psi, but you would need to consult a look-up table to discover this.  When any such a key is provided, whether transparent or not, we say that the instrument is calibrated.

 

The relationship between measuring states and object-properties, i.e., the measurement function, is semantic in exactly the same sense as demonstrated in the case of colour-vision in the last section.  That is, given that the instrument is in a particular state, the thing it is measuring appears, as far as the instrument goes, a certain way.  The calibrated notation on the face of the gauge gives us a way of expressing this property.  This connection persists even when a particular instrument is broken, out of range, or improperly connected.  Even when the gauge ceases to be a reliable indicator, it is still clear what description it yields. 

In the transparently notated gauge, it is only by recognizing that the instrument is in the ‘14’-pointing state that we get to say that the atmosphere is at 14 psi.  This is an epistemic connection, not a constitutive one.  That is, it makes perfectly good sense to allow that the atmosphere might not really be as the gauge indicates.  Even when the instrument is malfunctioning, the atmosphere continues to be at 14 psi, whatever the gauge might indicate.  In this sense, the state measured by an instrument transcends the state of the instrument itself.  This is the crucial feature of Dretske’s metaphor, and central to understanding the case of colour.[17]  It shows why it is not trivial to say, as we do in the Fundamental Principle, that something really is as the machine says it is when the machine is in state such-and-such.  This recognizes that there is more to the atmosphere being at 14 psi than an instrument indicating that it is so. 

The parallels between the measuring instrument and our sensory systems ought to help us understand the semantic character of sensory states.  However, there is a problem here.  Our sensory systems do not come marked in a convenient notation that we can use to describe the world; “this is not a courtesy that nature extends to us,” Dretske says (ibid., p. 47).   So how are we to concoct a notation that gives us a way of expressing their measurement functions? This is the problem Dretske tries to address when he asks us to imagine what would happen if we are given an instrument, as before, but find that “there is nothing there (or the numbers are no longer legible) to tell a curious onlooker what the pointer positions mean”.  This approximates the task of someone trying to discover what his own sensory systems measure.  Dretske says “if we know the instrument was working properly . . . one would simply determine, by independent means, what the pressure . . . is” (ibid., p. 48).   In other words, if the markings needed to translate a gauge-state into a description of the world are absent, we recalibrate the instrument.  If the unmarked instrument is connected to the atmosphere, and we find that the atmospheric pressure is 14 psi, then we can paint ‘14 psi’ on the face of the gauge right where the pointer is. 

The idea of recalibrating the gauge is disarmingly simple.  But it demands that collateral knowledge be available to the pressure gauge user, and puts him under too heavy an epistemic load.  How is the user to determine the pressure of the atmosphere in pounds per square inch simply by looking at states of an unmarked pressure gauge?  How is he even to know that it denotes pressure?  Yet, that, by analogy, is what the naïve perceiver is trying to do with regard to his own sensory systems.  A philosopher who represents the naïve perceiver’s sensory concepts in scientific notation exceeds the epistemic capabilities we can expect of unaided perception.  Similarly, “physicalist” specifications of colour properties, in terms of reflectance etc., fail properly to capture the content of colour vision as it presents itself to the naïve observer, to an animal, young child, or adult untutored in physics. 

 

B.  Auto-calibration

Consider now an approach to calibration epistemically less committed than Dretske’s.  Suppose that we paint arbitrary symbols on the blank face of the gauge – ‘A’, ‘B’, ‘C’, etc.  Each such mark helps us identify a measuring state of the gauge.  Now by analogy with the “transparent” gauge markings described in the last section, we can simply use the marks we have painted to express object properties.  Where before we said “The air is 14 psi” when the instrument is connected to the air and the needle pointed to ‘14’, now we say “The air is A” when it points to ‘A’, or “The tire is B” when it points to ‘B’.  These marks give us a way of identifying the instrument’s measuring-states, and at the same time, a way of describing objects connected to the instrument.  The marks on the instrument are used to designate object properties.  They constitute what I will call an auto-calibrated system of signs.  In such a system, easily accessible marks of an instrument’s measuring state are used to generate descriptions of the things that the instrument measures.  These descriptions express the same properties, but in different words – just as newtons per square metre can be used to express the same pressure properties as pounds per square inch, so ‘A’, ‘B’, ‘C’, etc. express the same pressure properties, but in a different notation.

Auto-calibrated signs give a user of the illegible gauge a way of describing the world.  And just as before, the properties described by these signs transcend the gauge-states themselves.  It still makes sense to say, in general, that the gauge is wrong.  (Suppose the needle points to ‘A’.  Then the gauge indicates that the thing being measured is A.  Now suppose that the user gives the gauge a sharp tap, and finds that it goes to ‘B’.  Now the user has evidence that the earlier reading was in error.)  When the needle points to A, the user has defeasible evidence that the measured object is A.  This evidence does not compel him to describe it in that way.  When he does in fact do so, he is endorsing what the gauge says.  So we could say: the air is A if it really is as the instrument makes it appear when it points to ‘A’.  This is analogous with the observation on which the Fundamental Principle of Colour Attribution rests: something has a particular colour-property if it really is the way it looks when it presents the look associated homophonically with that colour property. 

In order fully to understand an auto-calibrated sign, we need an explication of what it designates.  Now, there is a very simple way in which such signs can acquire meaning.  Imagine a naïve person using the auto-calibrated pressure gauge for various everyday purposes, for example, for checking tire and balloon pressures and the like.  The marks on the face of the gauge allow her to identify and compare readings on different occasions of the instrument’s use.  Thus, they furnish her with signs that she can use in generalizations.  For example: “When the bicycle tire is pumped up only to ‘A’, it goes bump when you ride over a curb, but when you pump it up to ‘B’, curbs are no problem”.  Or, in our auto-calibrated notation: “A tire at A will go bump, but a tire at B will not”.  These inductive generalizations are based on co-classifications – the gauge-user co-classifies things by means of the values of gauge markings registered by such objects, and uses such classifications as the basis for future induction.  My suggestion will be that auto-calibrated signs can acquire meaning by being associated with such a taxonomy.  With respect to these signs, meaning consists, as Wittgenstein insisted, in use.  (Note that if all such inductions fail, one might conclude that the instrument was not actually measuring the properties of the things to which it is connected.)

 

C.  Colour-Looks as Auto-calibrated Signs

The suggestion that I want to make is that colour-looks are auto-calibrated signs.  Just as the marks painted on the dial allow us to identify measuring-states of the gauge, so colour-looks are easily accessible marks by which states of the colour-vision system can be identified.  Like the ‘A’ on the dial, they also yield a notation for the things that this system “measures”.  When a perceiver S looks at a wooden tabletop, she is in visual state B.  This visual state B has a certain colour-look, say brown.  S uses this colour-look – this easily accessible feature of her own measuring state – to designate an object property of the object at which she is looking. 

What is the property designated by a colour-look?  I shall deal with this question in the next section.  (In addressing it, I shall be appealing, as I did at the end of the last sub-section, to the uses to which we instinctively put colour-information.)  What we have so far is an account of the way in which we use colour-looks as analogous to meaningful terms by which we designate properties.  Earlier, we found Dretske making the claim that our sensory systems do not come marked with a convenient notation which we can use to describe the objects they measure – this “is not a courtesy that nature extends to us,” he said.  We see now that this is false.  Colour-looks are auto-calibrated, and thus they are associated with a transparent system.  Note that this conception of a colour-look as an auto-calibrated sign accounts for the universal use of terms like ‘brown’ to denote both colour-looks and colour-concepts (cf. Peacocke [1984]).  The words we use to describe colour-looks carry over and become linguistic expressions for colour-properties by virtue of the “semantic” relationship between colour-looks and colour-properties. 

 

       IV.  COLOUR-PROPERTIES

 

A.  Task-oriented Taxonomies

Let us return now to Revelation.  What is the meaning of our attributions of colours like canary yellow to things?

I remarked in section II that there is a set of instinctive practices in which colour-experience enables us to participate.  Human visual systems instinctively use colours in at least the following ways:

 

(1) to co-classify things for purposes of induction, for example, to make generalizations by which one makes judgements concerning the ripeness of fruit, or the health of one’s conspecifics,

(2) to re-identify things on different occasions, for example, one’s car in a crowded parking lot,

(3) to segment the visual scene into figure and background,

(4) to find things by visual search, for instance red or orange fruit against a background of green foliage,[18] and

(5) to match and differentiate things by the colour-looks they present, in order to be able tell, for instance, which part of your uniformly coloured lawn is shaded by trees.

 

My thesis is that we instinctively use colour-looks in order to group things together for the above-mentioned tasks.  Colour-properties are equivalence concepts.  Things grouped together by colour are treated in the same way for the purposes mentioned above. 

Here is an example.  An induction base is a class of things which, because they share some feature, are expected to be similar in other ways – thus, when one member of an induction base is observed to have property F, this affects the subjective probability of as yet unobserved members of the induction base having F.  Colour-vision constructs induction bases using colour-looks.  Two things share a colour-property if they are assigned, because of colour-looks, to induction bases within which objects are expected to have further features in common.  They are different with respect to a colour-property if colour-vision assigns them to different induction bases corresponding to different expectations.  Colour categories form equivalence groupings not only for induction, but also for the other epistemic practices listed above.  Things of the same colour might tend to be assigned to the same figure against a ground of a different colour.  One uses colour when one is trying to decide which of several Toyotas in the parking lot is one’s own – it can only be so if it belongs to the same colour-equivalence class.  And so on. 

I do not mean that when things present the same look they are necessarily assigned to the same colour-classification, or that things are assigned to different colour-categories if they present different colour-looks.  Consider a case like this.  In the supermarket, a particular mango looks yellow.  I infer that it is ripe.  I take it home, and there it looks greenish.  Moreover, it turns out not to be ripe.  I conclude that in the supermarket, it looked different from the way it really was.  My assumption here is that though the mango presented different colour-looks in different conditions, it retained the same enduring colour-property.  And I might well conclude that in the supermarket, my senses told me something false; they assigned the mango to a category – the category I use for ripeness-inferences in mangos – to which it did not in fact belong.  In the supermarket, I attributed to the mango the colour-property designated by the yellow look it presented there.  At home, I come to realize that I was wrong to do so.  At home, I realized that it really has the property designated by the greenish look it presents there.  (Remember the Fundamental Principle of Colour Attribution.)  This shows that it is wrong to identify a colour with a fully specified look, or the colour a thing has now with the look it presents now.[19] 

What then is the significance of attributing canary-yellow to something? It is to say

 

first, that there is a region of colour space – the canary-yellow region – within which each minimal region designates a colour-property, and that canary-yellow is the union of all of these properties, and

 

second, that the colour-vision system instinctively groups things with this property together and treats them as similar for some of the above tasks. 

 

What is it to say that something is of a different colour, say lime-green?  That, lime-green is a property like the above, but that lime-green things are not equivalent to canary-yellow things with respect to any of the tasks mentioned above.  (There could be, of course, a broader colour category that includes both as sub-categories, but this broader category would be distinct from both canary-yellow and lime-green.) 

My proposal is that we should construe the meaning of colour-looks by reference to the ways in which we use them.  What do colour attributions say about external things?  Many philosophical theories of colour seek for the answer to this question by considering the information carried by colour-looks.  These theories concentrate on world-to-sensory-system links, in the hope that an adequate account of these will tell us what we can infer about the condition of the world from the states of our colour vision system.  However, because such inferences generally depend on information about distal conditions that is not available in colour experience alone, they end up offending against what Boghossian and Velleman call the “naïveté of vision”.  What I am suggesting here is that the colours should be defined in a different way, namely by means of the downstream connections between sensory-system states and the actions to which they are linked.  We know these connections implicitly and instinctively: we do not gather things together by learning and reasoning, but unreflectively.  Consequently, we know implicitly what our colour-attributions mean.  Since these downstream connections are triggered by colour-looks, it is reasonable to say that this knowledge is implicit in colour-experience.

The task-oriented conception of colour captures the intent of Revelation as weakened by the Empiricist Codicil in section I B above.  Since colour-properties are identified by colour-looks, it only takes experience “as of a canary-yellow thing” plus the task-oriented knowledge instinctively implicit in colour-vision to know the conditions under which two things are to be co-classified for purposes such as those mentioned above, and only this experience together with one as of a lime-green thing to know when canary-yellow things must be differentiated from lime-green things for those purposes. 

 

B.  Some Consequences

The task-oriented conception of colour implies a rather different approach to constancy and variability of colour-appearance than the problematic one sketched in section I.  Consider once again the case of human faces viewed under dense leaf cover.  Though colour photographs reveal that they reflect a greenish signal, they look pretty much the same colour as they always look.  The colour-realist takes this to mean that colour vision reveals an illumination-independent property of faces.  Thus, he takes such variability of colour-appearance as faces display to be an anomaly, a sign that colour vision is not perfectly well adapted to its purpose.  Under the task-oriented conception, constancy means simply that the epistemic practices fed by colour vision treat faces pretty much the same when they are under trees as when they are in unfiltered sunlight.  The sameness of these faces does have some role, no doubt, in explaining why they are so treated, much as the realist insists.  However, the task-oriented conception has no need for such a strong thesis; it can easily allow that the practices in question offer no information about co-classification in certain situations.  We need not trace this “failure” to an inability to detect an illumination-independent property: we can instead cite the nature of the practices in which we instinctively engage. 

This approach has three advantages.  First, it restricts itself to information available within the system.  What we know implicitly is that certain objects are reliably assigned to task-relevant equivalence classes only in a certain range of circumstances.  What we do not know implicitly is that they have the same property.  The former conception expresses our instinctive practice; the latter does not.  Consequently, task-oriented categories respect Revelation better than physical categories.  Second, task-oriented categories do not rely on information-based accounts of the adaptiveness of colour vision.  Such accounts are contentious: since colour vision may well have evolved in primates to aid visual search (for fruit among foliage), it may perform its task adequately if it heightens contrast between fruit and foliage, even though this might falsify the colours.  [20]  It may have no need, as long as it enhances this contrast, to respect properties that things have independently of us.  Finally, this approach is not obliged to treat colour variability as an anomaly.  As we have just seen, epistemic practices might work perfectly well with limited constancies. 

One implication of treating colours as equivalence concepts is that we will often be better at perceiving colour equivalences than at perceiving absolute colours.  This too seems to be a virtue of the account.  Consider the case of unsaturated colours.  Put a few drops of red into the white you use to paint your room, and your walls will look different at different times of the day.  The red-tincture will make it look pinkish in the reddish light of sunrise and sunset.  Now: is your wall really less pink than it looks at sunset?  Or is it really more pink than it looks at noon? Our sensory conceptions of colour offer us no hint as to how to answer such questions.  According to the approach that results in the Fundamental Principle, each look designates a particular colour property.  But we have no way of determining which (if either) of the two looks is correct.  Now, it might well be that we have highly predictable, or even innate, ways of discounting the variability of the wall’s colour appearance.  Thus, it might well be that vision informs us that the wall is the same colour at sunset as at noon.  However that may be, vision and its attendant mechanisms do not tell us what colour the wall really is, pinkish or white. 

The Fundamental Principle also accommodates a certain kind of pluralism with regard to colour concepts.  Imagine a wall obliquely illuminated from one side by sunlight coming through a window.  If the wall is white with a slight tincture of blue, one can imagine that the parts away from the window might present a bluish look, while the parts near the window look white.  Viewed as a whole, the wall might well look uniformly coloured, though its parts might look differently coloured if we were to look at them in isolation.  Now, suppose that we provide an observer with several samples of colour gradients that range from white to bluish, and ask him to match the wall with one of them.  He might well pick one of them, and thereby demonstrate that some part of his colour vision system does differentiate the colour-look presented by different parts of the wall.  This would show that we can have divergent colour attributions in different contexts, relative to different tasks.  Relative to the matching task that is subordinate to shadow detection, the wall is not uniform.  Relative to the induction task, it is uniform.  It is advisable, therefore, to be pluralistic about colour categories, not only about the colour categories used by different species, but also about the colour categories used by a single human organism.[21]

 

C.  The Superficiality of Colour

We are now ready to revisit the difference between musical harmony and colour.  Why is there information transfer from reality to experience in the case of harmony, but not in the case of colour?  The crucial point, I believe, is the difference in what I will call the ergonomic significance of the two modalities.  I have been arguing that in order properly to understand the significance of colour experiences, one needs to consider what one does with them as a matter of instinct.  The same is true for other sense experiences.  The difference between harmony and colour is, so I claim, that one of things that we do with harmony is manipulate our own voices to produce harmonies, but we do not know by instinct how physically to manipulate colour. 

The significance of musical perception is probably derived from the way in which we produce musical sounds.  A string produces sound by vibrating; if it is pegged at both ends, its amplitude will be zero at each end and maximal in the middle, and it will approximate a wave of length equal to the distance between the pegs.  Since the heard pitch of a sound depends on its wavelength, the primary tone that a pegged string produces will depend on its length (among many other factors).  Now, the main constraint on the vibration of a pegged string is that the amplitude must be zero at the pegs; all sorts of vibrations are possible in between, provided they are consistent with the tensile strength and elasticity of the string.  Thus strings produce subsidiary waves; the string will vibrate as a whole with the maximum amplitude at its mid-point, as described above, but it will also vibrate, at the same time, like two pegged strings with a zero point in the middle, and like three such strings, with two zero points evenly spaced along its length, and so on.  The lengths of these subsidiary waves, or overtones, will be in whole number ratios to the primary wave; generally speaking, the smaller the whole number, the more prominent the overtone.  The human voice produces sound by driving columns of air; these columns behave much like pegged strings.  This accounts for the timbre of the human voice: the musical sounds it produces are not pure, but a mixture of waves in (small) whole number ratios. 

Now, when we sing a note, and at the same time, somebody else sings a note that stands to ours in a whole number ratio, the second person’s sound resonates with and reinforces an overtone in our own voice.  This phenomenon gives us pleasure, and is at the centre of our appreciation of harmonies.[22]  Consequently, one of the ways in which humans can give one another pleasure is by singing together in the Pythagorean ratios.  Because this is so, musical expression depends not only on the capacity to hear these ratios in a particular way, but also on that of adjusting one’s own voice in such a way that it harmonizes with that of another.  These two things are closely linked.  I would contend that the auditory experience of musical harmonies is generally associated with the innate ability to adjust one’s own sound production to conform to external constraints.  Since this is so, there is some sort of connection between the physical structure of musical harmonies and the way in which we perceive them.  This, I speculate, is the reason why there is a transfer of structure from waveforms to auditory experiences. 

Colour experience is completely different in this respect.  The causal connections here are outside-in, but not inside-out; the experience of colour gives us information that enables us to undertake epistemic activities concerning external things, but it is not innately associated with the ability to produce or adjust the colour values of the things one sees.  (An artist can reproduce her colour experiences in paint.  But she has to go to Art School to learn enough about paints to do so.)  This is not because colour experience is “purely descriptive”, and lacks all significance for action.  It is, as we have seen, significant for epistemic actions.  The relevant point is that epistemic actions take place “inside the head”.  Sense experience in musical audition is associated with the ability to produce or influence external things, and hence it needs tacitly to contain (at least some) information about the physical character of external things.  Colour experience is not associated with any externally directed manipulation; hence, it does not need to contain information about the physical character of external things.  We noted in section I that the similarity space of colours is different from the similarity space of external properties like wavelength and reflectance.  This implies that the similarity that we experience as between two colours does not guarantee real similarity.  We have now explained why this can be so.

However, this does not completely dissolve the puzzle – it only tells us why there is no transfer from the structure of colour-reality to the structure of colour-appearance.  But induction and other epistemic practices are founded on similarity.  Why does induction based on colour work if the similarity on which it is founded is not real?  I would suggest that we could address this puzzle by noting that in fact nature is extremely sparing in the inductive inferences it attaches to colour.  Suppose that as an extremely naïve individual in a brand new culture you sample a canary yellow confection and find it pleasingly tart.  If there is another confection of the same colour on the same tray, you might well choose it, expecting that it too will be pleasingly tart in taste.  Later in the day, you are served a dish of rice that closely resembles the confection in colour.  Do you have any tendency at all to assume that the rice dish will be pleasingly tart?  I do not think so.  This reflects the conservatism of colour-based inductions in nature.  Birds identify edible fruit by colour, but they do not extend the edibility inference to other kinds of fruits or vegetation of the same colour, for instance to fruits of different shape or size.  If they did, they would likely be poisoned. 

Generally speaking, then, colour is a last differentia.  We tend not to make inferences of the form: “This Col thing is F, so all Col things are F.”  Rather we make inferences of the form: “This Col thing of kind K is F, so all Col things of kind K are F.”  In other words, it is only within specific kinds that we will allow colour to ground inductive generalizations.  If we were to make general inferences based on scarlet alone, it would be astounding if we came up with reliable results.  But we do not, and this increases the chances that our generalizations will be correct.  This defuses the puzzle.  [23] 


 

REFERENCES

Boghossian, Paul A., and J.  David Velleman

(1989) “Colour as a Secondary Quality”, Mind New Series 98: pp.  81-103.

(1991) “Physicalist Theories of Color”, Philosophical Review Vol.  100: pp.  67-106.

Bradley, Peter A.  and Tye, Michael (2001) “Of Colors, Kestrels, Caterpillars, and Leaves”, Journal of Philosophy 98: 469-87.

Cohen, Jonathan (2000) “Color Properties and Color Perception: A Functionalist Account”, Rutgers University PhD thesis.

Dretske, Fred (1995) Naturalizing the Mind, The Jean Nicod Lectures, 1995.  Cambridge Mass: MIT, Bradford Books. 

Hardin, C.  L (1988) Color for Philosophers: Unweaving the Rainbow Indianapolis: Hackett. 

Harvey, Jean (2000) “Colour-Dispositionalism and Its Recent Critics”, Philosophy and Phenomenological Research 61: 137-55.

Hurwich, Leo M.  (1981) Color Vision Sunderland MA: Sinauer.

Hurwich, Leo M.  and Jameson, Dorothea (1956) “Some Quantitative Aspects of an Opponent-Colors Theory.  IV.  A Psychological Color Specification Scheme”, Journal of the Optical Society of America 46: 416-21.

Jackson, Frank and Robert A.  Pargetter (1987)  “An Objectivist’s Guide to Subjectivism about Colour”, Revue Internationale de Philosophie Vol.  41: pp.  127-41.

Johnston, Mark

(1992)  “How to Speak of the Colors”, Philosophical Studies Vol.  68, pp.  221-63. 

(1997) (1992) reprinted with a Postscript in Readings on Color vol 1: The Philosophy of Color A.  Byrne and D.R.  Hilbert (eds), Cambridge Mass, pp.  137-76. 

Long, A.  A., and Sedley, D.  N.  (1987) The Hellenistic Philosophers vol 1 (Cambridge: Cambridge University Press).

Matthen, Mohan (1999) “The Disunity of Color”, Philosophical Review Vol 105: pp.  47-74

McLaughlin, Brian (forthcoming) “The Place of Color in Nature”, New Essays on Consciousness Quentin Smith (ed.), (Oxford: Oxford University Press).

Peacocke, Christopher (1984) “Colour Concepts and Colour Experience”, Synthese 58: 365-82.

Regan, B.C., Julliot, C., Simmen, B., Viénot, F., Charles-Dominique, P.  and Mollon J.D.  (2001) “Fruits, Foliage and the Evolution of Primate Colour Vision”, Philosophical Transactions of the Royal Society of London B 356: 229-83.

Tarski, Alfred (1944) “The Semantic Conception of Truth and the Foundations of Semantics”, Philosophy and Phenomenological Research 4: 341-76.


[1]   See Hurwich (1981) for bibliographical details.

[2] I use the term “suggests” deliberately; arguments from colour-constancy to colour-realism are far from conclusive.  Peter Bradley and Michael Tye (2001), recent defenders of colour-realism, seem to recognize this: “The fact that objects appear to retain the same color through a wide variety of changes (though certainly not all) strongly suggests that colors are illumination-independent properties of those objects” (p. 480).  They go on to appeal to “normal viewing conditions” to define the colour of a thing, but do not say how they will adjudicate which of several different but metaphysically equivalent sets of viewing conditions is “normal”. 

 

[3]  Johnston does not endorse this principle himself.  He cites Bertrand Russell as one philosopher who believed in Revelation.  The arguments of Boghossian and Velleman (1989, 1991) are based on something very close to this principle.

[4]  See Boghossian and Velleman (1991).

[5]  Paul Boghossian and David Velleman’s (1991) criticism of ‘physical­ist’ theories of colour rests in part on the idea that they fail these conditions.  They argue that if red and orange were categories of physics, then, since our per­ceiving some things as red and others as orange does not imply that these things belong to different physical categories, we might be mistaken about red and orange being different.  Boghossian and Velleman rightly insist that we cannot be mistaken about this.  They conclude that red and orange are not physical categories.

[6]  This needs some qualification, as Margaret Schabas tells me.  Because of the different overtones, or timbres, of various instruments, the same chord does not sound exactly the same on all of them.  Consequently, orchestras demand that certain instruments be played deliberately flat or sharp, so as to compensate for these variations, and thus to produce a coherent sound.  This shows that contrast phenomena affect our perception of the harmonies, much as they do in the case of colour.  The point that I am making is unaffected by this: the structure of our auditory experiences still reflects structure in the sounds we hear. 

[7]  The classic philosophically embedded account of this structure is C.  L.  Hardin (1988, chapter 1).

[8]  See Hurwich (1981), plate 1-5, and Hurwich and Jameson (1956) for quantitative details.

[9]  In fact, since colour-space is not granular, and indiscriminability is not transitive, an object will occupy several overlapping minimal regions of colour-space.  Thus, a colour-look should properly be identified as the set of minimal regions that contain a particular point in colour-space­. 

[10]  Jackson and Pargetter (1987), Cohen (2000), Harvey (2000), and McLaughlin (forth­coming) identify colour properties with colour-looks, and claim that the colour of an object changes with viewing conditions.  Or at least they think that there is an ontologically fundamen­tal class of colour-properties, “transitory colours” or “present colours”, which are the same as colour-looks.  They construct other, more enduring but ontologically derivative, class of colour-properties from colour-looks.  This “relativist” proposal entails that an observer cannot be wrong about the colour of an object in her particular viewing conditions, and I shall be arguing that this is false in general. 

[11]  In these principles, the colour-property attributed to the object of experience is distinguished from the colour-look along the lines of Peacocke (1984). 

[12]  Colour-looks can be specified by assigning ranges of values to the variables in the above predicate, as distinct from point-values.  The Dispositional Principle maintains that a thing is brown if this range is “standard”.

[13]  A more complex argument to the same conclusion can be found in Boghossian and Velleman (1989).

[14]  Cf.  Boghossian and Velleman (1989): “[S]urely, one can tell whether two objects appear similarly coloured on the basis of visual experience alone.  To be sure, one’s experience of the objects will not necessarily provide knowledge of the relation between their actual colours.  But the physicalist account implies that visual experience of objects fails to provide epistemic access, not just to their actual colour similarities, but to their apparent similarities as well.  And here the account must be mistaken.  The apparent colours of objects can be compared without empirical inquiry .  .  .” (p.  83, emphasis mine).

[15]  Sextus Empiricus attributes to Epicurus the theory that “True opinions are those attested and those uncontested by self-evidence.”  Long and Sedley (1987): 18 A (1), p.  90.  This is meant as a “criterion”, a theory that (a) tells us how to test an opinion for truth-value, and (b) defines truth in terms of a positive test outcome.

[16]   Dretske (1995), chapter 2.

[17]  Dretske’s metaphor might seem to beg the question in favour of realism, and against secondary qualities.  But this is not so: he doesn’t specify what kind of properties a measuring instrument might designate.  It might well designate a fully specified relational property in which the observer figures, i.e., a secondary property. 

[18]   See Regan et.  al.  (2001)

[19]   See note 10 for a list of philosophers who hold that the look a thing presents now is its “present colour”. 

[20]  See Regan et al (2001) for a recent review.  What is needed for fast search is the phenomenon known as “pop-out”, in which the time taken to search for a target is independent of the number of non-targets, or distractors, present in a scene.  To be effective for visual search, then, colour vision will need to exaggerate the difference between fruit and foliage to the point where pop-out occurs.

[21]  In Matthen (1999), I defended pluralism across species.  Retrospectively, it appears to me that this pluralism could be traced to the different tasks performed by different species of animals in their diverse habitats.  Since organisms perform a variety of tasks with their colour-vision, this same pluralism holds within an individual.

[22]  It should be noted that different cultures fasten on different aspects of musical sound; it is a characteristic of Western music that it emphasizes harmony.  But this does not mean that devotees of other kinds of music fail to appreciate harmony.

[23]  Jonathan Cohen, Tom Hurka, Ali Kazmi, Patrick Rysiew, and Catherine Wilson gave me much-needed aid in formulating the central theses of the paper.  Jill McIntosh has been an extraordinary editor, and I am grateful to her for many useful suggestions, both editorial and substantive.  This paper was delivered to the Bellingham Summer Philosophy Conference (2001), and to the Philosophy Departments of the Australian National University and the University of Calgary.  I thank these audiences for helpful discussion.