The Categorization of Colour

1 The Categorization of Colour As pointed out in the Preface, linguistics is concerned with categorization on two levels. In the first place, linguists need categories in order to describe the object of investigation. In this, linguists proceed just like practitioners of any other discipline. The noises that people make are categorized as linguistic or non-linguistic; linguistic noises are categorized as instances of a particular language, or of a dialect of a particular language; sentences are categorized as grammatical or ungrammatical; words are categorized as nouns and verbs; sound segments arc classified as vowels or consonants, stops or fricatives, and so on. But linguists are (or should be) concerned with categorization at another level. The things that linguists study—words, morphemes, syntactic structures, etc.—not only constitute categories in themselves, they also stand for categories. The phonetic form [jed] can not only be categorized as, variously, an English word, an adjective, a syllable with a consonant-vowel-consonant structure; [jed) also designates a range of physically and perceptually distinct properties of the real world (more precisely, a range of distinct visual sensations caused by the real-world properties), and assigns this range of properties to the category RED. The morphosyntactic category PAST TENSE (usually) categorizes states of affairs with respect to their anteriority to the moment of speaking; the preposition on (in some of its senses) categorizes the relationship between entities as one of contact, and so on. Both in its methodology and in its substance, then, linguistics is intimately concerned with categorization. The point has been made by Labov (1973: 342): 'If linguistics can be said to be any one thing it is the study of categories: that is, the study of how language translates meaning into sound through the categorization of reality into discrete units and sets of units.' Questions like: Do categories have any basis in the real world, or are they merely constructs of the human mind? What is their internal structure? How are categories learnt? How do people go about assigning entities to a category? What kinds of 2 77ie Categorization of Colour relationships exist amongst categories? must inevitably be of vital importance to linguists. Labov, in the passage just referred to, goes on to point out that categorization 'is such a fundamental and obvious part of linguistic activity that the properties of categories are normally assumed rather than studied'. In recent years, however, research in the cognitive sciences, especially cognitive psychology, has forced linguists to make explicit, and in some cases to rethink, their assumptions. In this first chapter, I will introduce some of the issues involved, taking as my cue the linguistic categorization of colour. 1.1 Why colour terms? There are good reasons for starting with colour terms. In many respects colour terminology provides an ideal testing ground for theories of categorization. It is commonly asserted—by linguists, anthropologists, and others—that categories have neither a real-world nor a perceptual base. Reality is merely a diffuse continuum, and our categorization of it is ultimately a matter of convention, i.e. of learning. This view was expressed very clearly by the anthropologist, Edmund Leach: I postulate that the physical and social environment of a young child is perceived as a continuum. It docs not contain any intrinsically separate 'things'. The child, in due course, is taught to impose upon this environment a kind of discriminating grid which serves to distinguish the world as being composed of a large number of separate things, each labelled with a name. This world is a representation of our language categories, not vice versa. Because my mother tongue is English, it seems self evident that bushes and trees are different kinds of things. I would not think this unless I had been taught that it was the case. (Leach 1964: 34) According to Leach, the categories that we perceive in the world are not objectively there. Rather, they have been forced upon us by the categories encoded in the language that we happen to have been brought up with. If categorization is language dependent, as Leach and many others suggest, it is only to be expected that different languages will encode different categorizations, none of them intrinsically any better founded, or more 'correct', than any other. Intuitively, we would probably want to reject, on common-sense grounds, the idea that all categories are merely learnt cultural artefacts, the product of our language, with no objective basis in Tlie Categorization of Colour 3 reality. Surely, the world does contain discrete nameable entities, and in many cases there does seem to be a natural basis for grouping these entities into discrete categories. Tables are one kind of thing, distinct from chairs; elephants are another, and quite different from giraffes. These cases need not concern us at the moment. There is, though, one area of experience where the reality-as-a-continuum hypothesis would seem to hold, and this is colour. It has been estimated that the human eye can discriminate no fewer than 7.5 million just noticeable colour differences (cf. Brown and Lenneberg 1954). This vast range of visible colours constitutes a three-dimensional continuum, defined by the parameters of hue (the wavelength of reflected light), luminosity (the amount of light reflected), and saturation (freedom from dilution with white). Because each of these dimensions constitutes a smooth continuum, there is no physical basis for the demarcation of discrete colour categories. Yet people do recognize discrete categories. It follows—so the argument goes—that these categories are a product of a learning experience, more particularly, of language. This view is supported by the fact that languages differ very considerably, both with regard to the number of colour terms they possess, and with regard to the denotational range of these terms. There are some well-known examples of non-correspondence of colour terms in different languages (see Lyons 1968: 56f). Russian has no word for blue; goluboy "light, pale blue" and siniy "dark, bright blue" are different colours, not different shades of the same colour. Brown has no single equivalent in French; the range of colours denoted by brown would be described in French as brim, matron, even jaune. Welsh glas translates into English as blue, green, or even grey. Very often, it is not just an individual colour term which does not have an exact equivalent in another language. Rather, it is the set of colour terms as a whole which fails to correspond with that of another language. Bantu languages are on the whole rather poor in colour terms; Tsonga, for instance, has only seven basic colour terms.1 These, with their approximate range of English equivalents, are as follows: (1) ntima: black rikuma: grey basa: white, beige 1 The notion of basic colour terms will be elaborated later, in s. 1.3. In addition to their basic colour terms, both Tsonga and Classical Latin (to be discussed below) have a large number of non-basic terms which denote quite precisely the colours characteristically associated with particular kinds of object. 4 The Categorization of Colour tshwuka: red, pink, purple xitshopana: yellow, orange rihlaza: green, blue ribungu: dark brown, dull yellowish-brown Tsonga divides the black-grey-white dimension in essentially the same way as English. However, only three categories are recognized in the hue dimension (tshwuka, xitshopana, rihlaza), whereas English has at least six (purple, red, orange, yellow, green, blue). Ribungu, on the other hand, is a special word for colours of low luminosity in the yellow-orange-brown region. Neither do we need go to non-European languages to find cases of extensive non-correspondence with English terms. Older European languages typically exhibit rather restricted colour vocabularies, which contrast strikingly with the modern English system. Consider the colour terms in Classical Latin (Andre 1949): (2) albus: white candidus: brilliant, bright white ater: black niger: shiny black ruber: red, pink, purple, orange, some shades of brown flavus: yellow, light brown, golden red viridis: green caeruleus: blue We find here, as in Tsonga, a rather restricted range of terms for the hue dimension. On the other hand, Latin made a distinction, lacking in English, between blacks and whites of high and low luminosity. Linguists have not been slow to recognize the theoretical significance of colour terminology. Consider the following passage from Bloomfield's classic volume Language: Physicists view the color-spectrum as a continuous scale of light-waves of different lengths, ranging from 40 to 72 hundred-thousandths of a millimetre, but languages mark olT different parts of this scale quite arbitrarily and without precise limits, in the meanings of such color-names as violet, blue, green, yellow, orange, red, and the color-names of different languages do not embrace the same gradations. (Bloomfield 1933: 140) This passage by Bloomfield could have been the model for Gleason's treatment of the same topic in his once very influential Introduction to Descriptive Linguistics: Tlie Categorization of Colour 5 Consider a rainbow or a spectrum from a prism. There is a continuous gradation of color from one end to the other. That is, at any point there is only a small difference in the colors immediately adjacent at cither side. Yet an American describing it will list the hues as red, orange, yellow, green, blue, purple, or something of the kind. The continuous gradation of color which exists in nature is represented in language by a scries of discrete categories— There is nothing inherent cither in the spectrum or the human perception of it which would compel its division in this way. The specific method of division is part of the structure of English. (Gleason 1955: 4). Other statements in the same vein could be quoted from other scholars. Indeed, many textbooks and surveys oflinguistic theory (the present work is no exception!) have an obligatory paragraph, even a whole section or chapter, devoted to colour. I would like to draw attention to one particularly important detail in the passage from Bloomfield, namely the assertion that colour categorization is arbitrary. Gleason, a few pages after the above quotation, makes the same point. What is more, Gleason puts his discussion of colour in the very first chapter of his textbook, as if to suggest that the arbitrariness of colour terms is paradigmatic for the arbitrariness of language as a whole. The arbitrariness of colour terms follows from the facts outlined above, namely the physical continuity of the colour space, and the human ability to make an incredibly large number of perceptual discriminations. There are, no doubt, other areas of experience which, like colour, constitute a smooth continuum: length, height, temperature, speed, perhaps even emotions like love, hatred, anger. Human beings can also make a large number of perceptual discriminations in these domains (but presumably nothing like the alleged 7.5 million colour discriminations). Languages are typically rather poor in their categorization of these domains. For length, English has only two terms, long and short. Colour, with its rich and language-specific terminology, is indeed an ideal hunting ground for anyone wishing to argue the arbitrariness of linguistic categories. 1.2 Arbitrariness Arbitrariness, as I have used the term in the preceding paragraph, has been a fundamental concept in twentieth-century linguistics. Its status as a quasi-technical term goes back to Saussure, who, in his Cours de 0 The Categorization of Colour Hnguistique generate (1916) proclaimed as a first principle of linguistic description that 'the linguistic sign is arbitrary': 'lc signe Hnguistique est arbitrage' (Saussure 1964: 100). The linguistic sign, for Saussure, is the association of a form (or signifier) with a meaning (or signified). There are two respects in which the linguistic sign is arbitrary (see Culler 1976:19ff.). In the first place, the association of a particular form with a particular meaning is arbitrary. There is no reason (other than convention) why the phonetic form [jed] should be associated with the meaning "red" in English; any other phonetic form, provided it was accepted by the generality of English speakers, would do equally well. It is therefore to be expected that different languages will associate quite different phonetic forms with a particular meaning; were the relationship not arbitrary, words with the same meaning in different languages would all have a recognizably similar form. With this characterization of arbitrariness, few would disagree.2 But there is another, more subtle aspect to arbitrariness, as Saussure conceived it. This is that the signified itself— the meaning associated with a linguistic form—is arbitrary. Saussure vigorously denied that there are pre-existing meanings (such as "red", "orange", etc.), which are there, independent of language, waiting to be named. The lexicon of a language is not simply a nomenclature for some universally valid inventory of concepts. There is no reason, therefore, why any portion of the colour space should have a privileged status for categorization in the colour vocabulary of a language; indeed, strictly speaking, there is no reason why colour should be lexicalized at all. We return, then, to the theme of Section 1.1. Reality is a diffuse continuum, and our categorization of it is merely an artefact of culture and language. The arbitrariness of the linguistic sign is closely linked to another Saussurian principle, namely the notion of language as a selfcontained, autonomous system. 'La langue', according to Saussure, 'est un systeme dont tous les termes sont solidaires et oil la valeur de l'un ne resultc que de la presence simultanee des autres' (1964: 159). The meaning of a linguistic sign is not a fixed property of the linguistic sign considered in and of itself; rather, meaning is a function of the value of the sign within the sign system which constitutes a language. : The doctrine of the arbitrariness of the signifier-signified relationship disregards, of course, the relatively rare phenomena of onomatopoeia and sound symbolism. It is worth mentioning that Rhodes and Lawler (1981) have recently suggested that the phonetic motivation of the signifier might be much more extensive than is traditionally believed. Vie Categorization of Colour 7 Thus concepts, i.e. the values associated with linguistic signs, arc purely differential; they are defined 'non pas positivement par leur contenu, mais negativement par leurs rapports avec les autres termes du systeme' (p. 162). This means that while the word red is obviously used by speakers of the language to refer to properties of the world, and might well evoke in the mind of a speaker a mental image of the concept "red", the meaning of the word is not given by any properties of the world, nor does it reflect any act of non-linguistic cognition on the part of a speaker. The meaning of red results from the value of the word within the system (more precisely, the subsystem) of English colour vocabulary. The fact that English possesses words like orange, pink, and purple effectively limits the denotational range of red in contrast with, say, Tsonga, which has only one word for the rcd-pinkpurple area of the spectrum. Should English acquire a new colour term, or should one of the existing colour terms fall into disuse, the whole subsystem would change, and each term in the subsystem would acquire a new value. There arc a number of implications for the study of colour terms which follow from the structuralist approach to word meaning. Amongst these are the following: (a) All colour terms in a system have equal status. Some colour terms might be used more frequently than others, but since the value of any one term is determined by its relation to all the other terms in the system, no one term can have a privileged status. (6) All referents of a colour term have equal status. Admittedly, the structuralist view does allow for the possibility of boundary colours. Recall the earlier quotation from Bloomfield, in which he stated that languages mark off different parts of the colour space 'without precise limits'. There will be regions between adjacent colour categories where unambiguous categorization will be difficult. Discounting such marginal cases, the structuralist view assigns to each exemplar of a colour category equal status within that category. If two colours are both categorized as red, i.e. as the same colour, linguistically speaking, then there is no sense in which one is redder than the other. This does not mean, of course, that an English speaker cannot perceive any difference between the two colours; only that for the purposes of linguistic categorization the difference is ignored. (c) The only legitimate object of linguistic study is the language system, not individual terms in a system. Neither can one legitimately 8 Vie Categorization of Colour compare single lexical items across different languages. Rather, one must compare entire systems, and the values of the items within those systems. 1.3 An alternative approach: focal colours In Sections 1.1 and 1.2 I have tried to give as objective and sympathetic an account as possible of the structuralist approach to colour terminology. I now want to present some arguments against the structuralist view. The pioneering work in this regard is Basic Color Terms (1969), by the linguist-anthropologists Berlin and Kay. On the basis of an investigation of the colour terms in ninety-eight languages, Berlin and Kay state: Our results ... cast doubt on the commonly held belief that each language segments the three-dimensional color continuum arbitrarily and independently of each other language. It appears now that, although different languages encode in their vocabularies different numbers of basic color categories, a total universal inventory of exactly eleven basic color categories exists from which the eleven or fewer basic color terms of any language arc always drawn. (Berlin and Kay 1969: 2) Berlin and Kay restricted their investigation to what they called basic colour terms. I shall have more to say about basic level terms in Chapter 3. Here, we can content ourselves with Berlin and Kay's operational definition. Amongst the characteristics of basic colour terms, as understood by Berlin and Kay, are the following. Basic colour terms (a) are not subsumed under other terms. Crimson and scarlet are not basic terms in English, since they are varieties of red. Orange is a basic term, since it is not subordinate to any other colour term; (b) are morphologically simple. Terms like bluish, bluish-green and chocolate-coloured, even golden, are excluded; (c) are not collocationally restricted. Blond, which describes only hair, is not a basic colour term; (d) are of frequent use. Rare words like puce, and technical words like xanthic, are excluded.3 ' It might be observed that these 4 criteria do not necessarily give unambiguous results. For some speakers, terms like mauve, lavender, lime, burgundy seem to have Vie Categorization of Colour 9 Berlin and Kay make two especially interesting claims. The first concerns so-called 'focal' colours. If people of different language backgrounds are shown a colour chart or an array of colour chips and are asked to trace the boundaries of the colour terms in their respective languages, one gets an impression of enormous crosslanguage variability (as well as of considerable variability between speakers of the same language; even the same speaker might perform differently on different occasions). Thus, two colour samples might well be categorized as the same by speakers of one language, but as different by speakers of another. If, on the other hand, people are asked to select good examples of the basic colour terms in their language, cross-language (and within-language) variability largely disappears. Although the range of colours that are designated by red (or its equivalent in other languages) might vary from person to person, there is a remarkable unanimity on what constitutes a good red. Paying attention to the denotational range of colour terms highlights the language specificity of colour terminology; eliciting good examples of colour terms highlights what is common between languages. By studying the focal reference of basic colour terms, Berlin and Kay were able to make their second, and somewhat more controversial claim. They noted that the ninety-eight languages in their survey appeared to select their basic colour terms from an inventory of only eleven focal colours. Furthermore, the languages did not select randomly from this inventory. If a language has only two colour terms (no language, apparently, has fewer than two), these will designate focal black and focal white. If there is a third term, this will always be red. The fourth term will be either yellow or green, while the fifth will be the other member of the pair yellow and green. The sixth term will be blue, and the seventh, brown. The remaining four colours (grey, orange, purple, and pink) do not show any special ordering. These generalizations may be expressed in the form of an implicational hierarchy: basic level status, for others not. Interesting in this connection is Robin LakofTs (1975) claim thai women tend to employ a more precise and more differentiated colour vocabulary than men. If this claim is true, women might in general possess a larger number of basic colour terms than men. Vie Categorization of Colour 10 (3) black < red < white grey orange yellow < green blue < brown < purple pink (3) is to be interpreted as follows: the existence in a language of a category to the right of an arrow implies the existence of all the categories to its left; the reverse implication does not necessarily hold. If a language has a colour term designating, say, focal blue, we can predict that the language will also possess the five colour terms to the left of blue; we cannot, however, predict whether it will have the colour terms to the right. Both in its methodology and substance, Berlin and Kay's work is not immune to criticism; see, for example, McNeill (1972) and Sampson (19806: 96ff). For twenty of the languages investigated, Berlin and Kay had access to bilingual informants who happened to be available in the San Francisco region. The responses of these informants could well have been influenced by their knowledge of English and by their exposure to a technological culture. Even more suspect are the data for the remaining seventy-eight languages in the survey; these were gleaned from dictionaries, anthropologists' reports (some dating from the last century), and oral reports from field-workers. No doubt, these deficiencies are part of the price one has to pay for a study of such breadth and generality as Berlin and Kay's. Even so, before we can discuss the linguistic implications of Berlin and Kay's work, it is necessary to see whether their basic insights concerning focal colours stand up to more rigorous experimental testing. With this in mind, let us turn to the work on colour terms conducted in the early 1970s by the cognitive psychologist Eleanor Rosch (published under her former name, Eleanor Heider). Heider (1972) reports four experiments which both confirm and elaborate some of Berlin and Kay's claims. The first experiment tested the stability of focal colours across languages. It was found that when subjects from eleven different language backgrounds were asked to pick out good examples of the colour terms in their respective languages, there was indeed a high degree of agreement concerning which colours were selected. When asked to point to a good example of red (or its equivalent in other languages), subjects tended to pick out Tlie Categorization of Colour 11 the same shade, irrespective of which language they spoke. The second experiment investigated some of the behavioural correlates of colour focality. Subjects from twenty-three language backgrounds were presented with samples of focal and non-focal colours, which they were asked to name. Subjects responded in their native language, and it was found that focal colours were named more rapidly, and that the names given to focal colours were shorter (when written out, the names contained fewer letters), than was the case with non-focal colours. This strongly suggests that focal colours are perceptually and cognitively more salient than non-focal colours. Experiment three was a short-term memory task. Subjects were shown a colour sample for a period of five seconds. Then, after an interval of thirty seconds, they had to identify from an array of colours the colour that they had just seen. At issue was whether focal colours would be recognized more rapidly and more accurately than non-focal colours. Two groups of subjects participated in this experiment. One group consisted of twenty native speakers of English. The other was made up of twentyone monolingual speakers of Dani. The Dani are a Stone Age people of New Guinea, whose language is one of the very few in the world which have only two colour terms. Between them, these two terms categorize the whole of the colour space, mola referring both to focal white and to warm colours (red, orange, yellow, pink, purple), while mili designates focal black and cool colours (blue, green). It was found that, overall, the English speakers could recognize the colours they had seen more accurately than the Dani. This suggests that colour memory is indeed aided by the existence of the relevant colour terms in one's language. (Another possibility, of course, is that a Stone Age culture, in which such things as traffic lights and colour-coded electric wires are unknown, provides little practice and few incentives for the memorization of colours.) More interesting was the finding that although the Dani's overall performance was poorer than that of the English speakers, they nevertheless performed better on the focal than on the non-focal colours. In this respect, the Dani did not differ from the English speakers. This aspect of the Dani's performance could not have been a consequence of the greater codability of focal colours, since the subjects did not possess separate lexical items in their language for designating these colours. Additional support for this view comes from Heider's fourth experiment. Here, Dani speakers were tested for long-term colour memory in a paired-association learning task. As expected, the subjects learned names for focal colours 12 The Categorization of Colour faster than names for non-focal colours. Further evidence for the perceptual and cognitive salience of focal colours comes from Heider (1971). Here it was found that three-year-old children, who had not yet acquired the full range of English colour terms, were more attentive to focal colours than to non-focal colours; also, three- and four-year-olds were able to match focal colours better than non-focal colours. The evidence for Berlin and Kay's other claim, concerning the implicational hierarchy of focal colours, is less robust. (3) suggests that while focal colours as a whole have greater perceptual and cognitive salience than non-focal colours, some focal colours (namely those on the left of the hierarchy) are more salient than others. Heider addressed this issue in one of the experiments already referred to. In the second experiment reported in Heider (1972), focal colours could be named more rapidly than non-focal colours. There were also differences in the speed with which focal colours could be named. Black was named most rapidly of all, followed by (in order of increasing delay) yellow, white, purple, blue, red, pink, brown, green, and orange. This ordering does not correlate significantly with the ordering of the colour terms in (3), neither was there any significant correlation between the implicational hierarchy and the relative salience of focal colours for the three- and four-year-old children studied in Heider (1971). Indirect evidence, whose significance is however difficult to evaluate, for the implicational hierarchy may be sought in other places. Position on the hierarchy tends to correlate with the productivity of certain derivational processes. Only terms at the very left of the hierarchy undergo derivation by means of the causativeinchoative suffix -en: whiten, blacken, redden (cf. 'bluen, 'yellowen, 'pinken, etc.); terms at the very right do not readily form abstract nouns in -ness: *purpleness, 'orangeness (cf. whiteness, blueness, greyness). We also note a weak correlation between position on the hierarchy and frequency of usage. Data in Kucera and Francis (1967) give black and white as the most frequently used terms, followed, in order of decreasing frequency, by red, brown, blue, green, grey, yellow, pink, orange, and purple. The empirical claims made by Berlin and Kay (1965) with regard to the implicational hierarchy are probably too strong. Firstly, the proposal that all languages in the world select from a universal inventory of just eleven focal colours needs relaxing. Russian, with words for light and dark blue, has twelve basic level terms. Arguably, The Categorization of Colour 13 some English speakers too have additional basic level terms {mauve, turquoise, etc.). We can also find languages whose inventory of colour terms does not conform to (3). Languages which do not have separate terms for blue and green, but which nevertheless have terms to the right of blue, are by no means infrequent. As may be seen from (1), Tsonga, with a term for grey, fails to conform. The same is true of Zulu. Zulu, like most Bantu languages, does not distinguish between green and blue, yet the language possesses a term for focal brown, nsundu. Interestingly, however, terms for green-blue—a category which Kay and McDaniel (1978) call 'grue'—often turn out to be bifocal, that is to say, the grue term refers both to focal blue and to focal green, rather than to one or the other of the two focal colours (or to an in-between colour). Certainly, Zulu speakers think of blue and green as different colours, and, if necessary, distinguish them formally by means of the expressions luhlaza njengesibhakabhaka "grue like the sky" and luhlaza njengotshani "grue like the grass". The years following the publication of Basic Color Terms saw a great deal of research on colour terminology (for a review, see Bornstein 1975). This led, amongst other things, to modifications of the implicational hierarchy (Kay 1975; Kay and McDaniel 1978). The details need not concern us here. Suffice it to say that this body of colour research presents a serious challenge to the structuralist approach to colour terminology. It is not that Berlin and Kay, or subsequent researchers, attempted to minimize the sometimes very different denotational ranges of colour terms in different languages, nor did anyone take issue with the notion of colour space as a physical continuum. But a factor was introduced which the structuralists had ignored, namely perception. It will be recalled that Gleason, in the passage cited earlier, explicitly stated that 'there is nothing inherent either in the spectrum or the human perception of it which would compel its division' (Gleason 1955: 4; emphasis added). At least since the researches of Helmholz, in the middle of the last century, it has been known that colour perception begins in the retina, with the stimulation of light-sensitive cells known as rods and cones. There are three kinds of cone. These react selectively to light in the red, green, and blue regions, while the rods are activated by the brightness dimension. More recent research has studied colour processing beyond the retina. (For a summary and discussion of the implications for colour terminology, see Kay and McDaniel 1978; von Wattenwyl and Zollinger 1979). It seems that green and red, and yellow and blue, 14 Vie Categorization of Colour stimulate complementary patterns of cell responses in the neural pathways between the retina and the brain. So, while it may be valid to talk of the colour spectrum as a smooth continuum, it does not follow that perception of the spectrum is equally smooth. From a perceptual point of view, it certainly does make sense to speak of an optimum red. An optimum red would be light of a wavelength which produces a maximum rate of firing in those cells which are responsive to light in the red region. Gleason, Bloomfield, and others not only leave the physiological basis of colour perception out of account, they also ignore environmental factors. Colour perception is not only a function of properties of the light waves entering the eye (Miller and Johnson-Laird 1976: 336). Just as objects are perceived to retain a constant size and shape, irrespective of their location and orientation with respect to the viewer, so the human visual system normalizes variations in the visual stimulus caused by changes in illumination of the perceived object. It might well be valid, at a certain level of theoretical abstraction, to speak of colour as a three-dimensional space. But people do not encounter colours as points in mathematical space, colours come as relatively stable properties of things. It is only in comparatively recent times, and only in technologically advanced societies, that it has been possible for a vast range of diverse colours to be applied, through industrial processing, to things. In the world of nature, things are typically associated with quite narrow segments of the colour continuum. Blood is, within a rather narrow range, red, milk is white, charcoal is black, lemons are yellow. By reversing the terms on either side of the copula, we obtain ostensive definitions of the colours: red is the colour of blood, white is the colour of milk, and so on. (Note, by the way, that many colour terms, in English and other languages, were originally names for objects. Examples from English include pink and orange, as well as violet, burgundy, and lime) Also from an ecological point of view, then, it is not really surprising that colour terms should refer, primarily, to rather restricted portions of the spectrum. Equally, the cross-language stability of colour focality may well have as much to do with the stability of the attributes of certain kinds of things, as with neurological processes of perception (cf. Wierzbicka 19806:42 f). It is along these lines, also, that we might attempt to explain the highly puzzling merging of blue and green in many languages of the world/1 4 Such languages arc particularly frequent in Africa and the Americas, and examples have been reported from Europe. Certain conservative dialects of southern Italy, for Vie Categorization of Colour 15 Why is it that just these colour categories should coalesce into a bifocal category? Blue is, of course, the colour of the sky, and green is the colour of grass. Yet unlike the red of blood and the yellow of lemons, the blue of the sky and the green of grass are highly variable; furthermore, the sky is not a tangible object whose surface can be touched. Blue and green thus lack the referential stability which nature provides for other focal colours, a fact which may go some way towards explaining the somewhat special status of these two categories. Given the focality of colour categories—whether this be the consequence of neurological processes of perception, of environmental factors, or of both—the structuralist account of colour terminology turns out to be grossly inadequate. Two characteristics of colour terms, in particular, are at variance with the assumptions of structuralism: (a) Colour categories have a centre and a periphery. This means that, contrary to structuralist principles, members of a category do not all have the same status. A colour term denotes, first and foremost, a focal colour, and it is only through 'generalization from focal exemplars' (Heider 1971: 455) that colour terms acquire their full denotational range. Obviously, if a language has relatively few colour terms, the denotational range of each term could well expand to take in a relatively large portion of the colour space. The centre, however, will remain constant. (b) Because of the primacy of focal reference, colour terms do not form a system, in the Saussurian sense. The focal reference of a colour term, e.g. red, is independent of whether yellow, orange, purple, etc., are lexicalized in the language. The addition of a new term, such as orange, might cause the total denotational range of red to contract, but the centre of the category will remain unchanged. In brief, colour terminology turns out to be much less arbitrary than the structuralists maintained. Colour, far from being ideally suited to demonstrating the arbitrariness of linguistic categories, is instead 'a prime example of the influence of underlying perceptual-cognitive [and perhaps also environmental: J.T.] factors on the formation and reference of linguistic categories' (Heider 1971: 447). instance, lack a term lor blue, verde (or its cognates) serving for both blue and green (Kristol 1980). 16 The Categorization of Colour 1.4 Autonomous linguistics vs. cognitive linguistics In this chapter I have outlined two radically divergent approaches to colour terminology. Although we have been concerned with a minute segment of any one language (we are dealing with, at most, a dozen or so words in any one case), the two approaches are symptomatic of two equally divergent conceptions of the nature of language. The contrast between the two conceptions will, in its various guises, constitute one of the themes of this book. At this point, therefore, it would be appropriate to highlight the basic issues dividing the two approaches. Structuralism maintained that the meaning of a linguistic form is determined by the language system itself. The world out there and how people interact with it, how they perceive and conceptualize it, are, in the structuralist view, extra-linguistic factors which do not impinge on the language system itself. Of course, people use language to talk about, to interpret, and to manipulate the world, but language remains a self-contained system, with its own structure, its own constitutive principles, its own dynamics. Language, in a word, is autonomous. With the advent of Chomsky's generative-transformational paradigm, the notion of the autonomy of language acquired a rather different sense. In the first place, language was no longer regarded as a self-contained system, independent of its users; rather, the object of investigation is a 'system of knowledge' (Chomsky 1986: 24) residing in a person's brain. In Chomsky's work, this mentalistic conception of language (which the present writer fully endorses) goes with the much more controversial claim of the modularity of mind: 'What is currently understood even in a limited way seems to me to indicate that the mind is a highly differentiated structure, with quite distinct subsystems' (Chomsky 1980:27). Just as the human body consists of various parts, each with its own function and developmental history, so the human mind consists of components which, though interacting, nevertheless develop and operate independently. One such component is the language faculty. The language faculty is viewed as a computational device which generates the sentences of a language through the recursive operation of rules on structured strings of symbols, assigning to each sentence thus generated a phonetic representation and a semantic interpretation, or logical form. It is the language faculty, thus understood, which determines a person's grammatical competence, i.e. linguistic competence in the narrow sense. Language is autonomous in Tlic Categorization of Colour 17 the sense that the language faculty itself is an autonomous component of mind, in principle independent of other mental faculties. The main concern of linguistics, in the Chomskyan mould, is the study of grammatical competence, i.e. the strictly linguistic knowledge which a speaker has acquired in virtue of the properties of the language faculty. As Chomsky is well aware, one can only maintain the thesis of the autonomy of language at the cost of extreme idealization: The actual systems called 'languages' in ordinary discourse are undoubtedly not "languages' in the sense of our idealizations. ... [They] might ... be •impure' in the sense that they incorporate elements derived by faculties other than the language faculty. (Chomsky 1980: 28) The 'impurity' of actual languages results from the interaction of the language faculty proper with at least two other components of mind, pragmatic competence and the conceptual system. The former has to do with 'knowledge of conditions and manner of appropriate use, in conformity with various purposes' (Chomsky 1980: 224). If grammatical competence characterizes the tool, pragmatic competence as it were determines how the tool is to be put to use. The conceptual system, on the other hand, has to do with matters of knowledge and belief; it permits us to 'perceive, and categorize, and symbolize, maybe even to reason in an elementary way' (Chomsky 1982: 20). It is, in Chomsky's view, the yoking of the conceptual system with the computational resources of language faculty that gives human language its rich expressive power and which makes human language qualitatively different from animal communication systems. Where, in the Chomskyan scheme, do the facts of colour categorization that we have considered in this chapter belong, to the conceptual system, or to the language faculty? Presumably, the answer most in keeping with the doctrine of modularity is: the conceptual system. This answer implies that the meanings of colour terms in a language are not, in effect, facts of language at all, in the narrow sense. Language, as a computational system for generating sentences, has nothing to do with how a person conceptualizes his world, how he perceives it and how he interacts with it.5 The issue is by no means so clear-cut, however. Especially in his more recent writings, Chomsky allows for the possibility that 'the state of knowledge attained may itself include ! Here, and elsewhere in the text, "he' is used as a 3rd-person pronoun unmarked for gender. 18 "Die Categorization of Colour some kind of reference to the social nature of language' (Chomsky 1986: 18). He has also conceded that it is not always an easy matter to distinguish between 'intrinsic meanings', i.e. meanings assigned by the operation of grammatical competence alone, and the interpretation given to sentences on the basis of beliefs about the world: Knowledge of language is intimately related to other systems of knowledge and belief. When we identify and name an object, we tacitly assume that it will obey natural laws. It will not suddenly disappear, turn into something else, or behave in some other 'unnatural' way; if it does, we might conclude that we have misidentified and misnamed it. It is no easy matter to determine how our beliefs about the world of objects relate to the assignment of meanings to expressions. Indeed, it has often been argued that no principled distinction can be drawn. (Chomsky 1980: 225) Chomsky (1986: 18) states that the blurring of the distinction between the purely linguistic and non-linguistic components of language knowledge does not give rise to 'conflicts of principle or practice' for proponents of the modularity hypothesis. In this book, I shall take the reverse position, i.e. that no distinction needs to be drawn between linguistic and non-linguistic knowledge. The facts of colour categorization as manifested in the meanings of colour terms are at once both facts about human cognition and about human language. Informing the content of the following chapters will be a conception of language as a non-autonomous system, which hypothesizes an intimate, dialectic relationship between language on the one hand and more general cognitive faculties on the other, and which places language in the context of man's interaction with his environment and with others of his species. On this view, a clean division between linguistic and non-linguistic faculties, between linguistic facts and non-linguistic facts, between a speaker's linguistic knowledge proper and his non-linguistic knowledge, between competence and performance, may ultimately prove to be both unrealistic and misleading. Criticism of the autonomy hypothesis, both in its structuralist and generative-transformational guises, is not new. More than half a century ago, Malinowski wrote: Can we treat language as an independent subject of study? Is there a legitimate science of words alone, of phonetics, grammar and lexicography? Or must all study of speaking lead to the treatment of linguistics as a branch of the general science of culture? ... The distinction between language and speech, still supported by such writers as Buhler and Gardiner, but dating back to De Tlie Categorization of Colour L9 Saussure and Wegener, will have to be dropped. Language cannot remain an independent and self-contained subject of study. (Malinowski 1937: 172) The same point has been made by George Lakoff (1978: 274), who maintains that it is unrealistic to speak of a language faculty independent of 'sensory-motor and cognitive development, perception, memory, attention, social interaction, personality and other aspects of experience'. In recent years, a number of linguists who are sceptical of the autonomy hypothesis, who believe, with Lakoff, that aspects of experience and cognition are crucially implicated in the structure and functioning of language, have given the term 'cognitive' to their approach. With the publication in 1987 of two monumental monographs—Langacker's Foundations of Cognitive Grammar and Lakoff s Women, Fire, and Dangerous Tilings—the approach is likely to exert an increasing influence on the direction of linguistic research for some years to come. This said, it should not be forgotten that the cognitive approach is much older than the work of the self-styled cognitive linguists. Scholars standing outside the mainstream of autonomous linguistics, whether structuralist or generative, have frequently worked on assumptions which present-day cognitive linguists would readily support. Important in this respect is Geeraerts's recent reappraisal of the now largely ignored work of the great European historical philologists (Geeraerts 1988a). Cognitivist assumptions also inform the work of many present-day linguists who, in spite of large differences in oudook, nevertheless search for an explanation of language structure outside a narrowly defined language faculty. Important contributions have been made by Jackendoff (1983), Hudson (1984), Wierzbicka (1985), and Givon (1979), as well as by a number of researchers into language acquisition (Slobin, Schlesinger, and others). The aim of this book is to explore some aspects of linguistic categorization, given the assumptions of the cognitive approach. Probably little of what I will have to say can be construed as decisive evidence against the autonomy hypothesis and the modularity hypothesis to which it is related. As Botha (1987) has wittily shown, the imposing intellectual construct erected by Chomsky is in a very real sense impregnable. Evidence on how people categorize the world through language can always be shunted off to the non-language faculties of the mind or dismissed as instances of the 'impurity' of 20 The Categorization of Colour 'actual languages', with little consequence for the autonomy of the language faculty proper. Yet differences in approach are real enough. Given the theoretical question of whether language behaviour needs to be explained in terms of a purely linguistic faculty, or whether language behaviour follows from more general cognitive abilities, the natural starting-point, as Lakoff (1977) has pointed out, is surely the null hypothesis, i.e. the assumption that there are no purely linguistic abilities at all. Only when the null hypothesis has been shown to be inadequate does the need arise to posit language-specific principles. Hopefully, the following chapters will show, not perhaps that the null hypothesis is fully adequate, but that it does, at least, permit a coherent account of a wide range of linguistic phenomena.