Colour vocabulary
At first blush, the domain of colour seems a prime example of language introducing distinctions within the continuous flux of our experience. It has been known for a long time that different languages divide up the visible spectrum in various and incompatible ways, and the consequent difficulty of translating colour terms from one language to another has often been commented on. There often is no simple equivalent for one language’s colour words in another: Ancient Greek khlōros is not really green, its typical English translation, but a green-yellow colour combination which can refer to the colours of young grass/leaves, sand and honey (Liddell and Scott 1940: 890). Since we all see the same wavelengths of light, this cross-linguistic variation cannot be the result of differences in colour perception, but must be language-internal: we see the same col ours, but different languages name them differently. For this reason, colour naming was often taken as the crucial example of the linguistic relativity hypothesis (see 11.5). The cross-linguistic variation in colour naming strategies suggests that speakers do not simply label pre-existing categories, which are objectively out there, ‘staring them in the face’. Instead, languages create the colour categories for themselves and, once created, speakers categorize colours in the terms particular to their language. Colour categories, in other words, are not universal, but language relative.
However, a tradition of research inaugurated by Berlin and Kay (1969) and still continuing has led to some important revisions of this relativist picture. Berlin and Kay hypothesized that each language has a set of basic colour terms (BCTs). In English, black, white, red, yellow, green, blue, purple, and grey are BCTs, whereas violet, ochre, eggshell blue, turquoise, etc. are not. To be counted as a BCT, a colour term has to meet the following criteria:
• it must be psychologically salient, i.e. one of the commonly remembered and used terms. This excludes English russet, cyan, jade, sepia, magenta, etc.;
• it must not be included in the range of another colour term; this excludes crimson (a type of red), indigo (a type of purple), tan (a type of brown) and light green (a type of green);
• it must be morphologically simple. This excludes grey-green and reddish;
• it must be applicable in all contexts, i.e. able to refer to a colour regardless of the object which has it. This excludes English sorrel and blonde, which primarily refer to colours of horses and hair/furniture/ beer respectively.
The number of BCTs can vary widely from language to language. The smallest number recorded is just two; on the other side, very few languages have more than eleven. Berlin and Kay and their colleagues have now explored the typical range of reference of BCTs in a wide range of different languages. They did this using 330 colours from the Munsell colour system, a standardized set of samples showing fi ne gradations between colours, rather like the colour sample cards available in paint shops. They particularly concentrated on what they termed the focus of colour terms. This meant the best example of a particular colour. For instance, the focus of English red, ‘focal red’, is that particular shade of red (say London bus red) which speakers would indicate as the ‘reddest’ red possible.
QUESTION What is the full set of ‘basic colour terms’ of English, according to the criteria listed above?
The general findings of work in the Berlin and Kay tradition challenges relativist beliefs about colour terms. Berlin and Kay discovered that even though languages differed in the number of their colour terms, and in the boundaries of any one of their terms, speakers show remarkable con vergence, both between and within languages, in the particular shades they nominate as the focal colours of each category. Out of the 330 Munsell colours Berlin and Kay presented to speakers, only a pool of thirty closely similar colours were chosen as examples of focal hues. These thirty were concentrated on the most typical examples of black, white, red, green, yellow, blue, grey, brown, orange, purple and pink. Take as an example those languages with a term for ‘red’. There is, in general, wide variation in the range of colours which speakers of these languages will count as examples of this category: sometimes it includes pale, pinkish reds, sometimes oranges, sometimes even whites. But in spite of this wide range of reference, there is consistent agreement between speakers of different languages about what counts as the best example of red: speakers consistently choose a very narrow range of Munsell chips, regardless of how many other shades the term also covers. Similarly, some languages have a term for ‘blue/green’, but speakers always chose either a single shade of blue (focal blue) or a single green (focal green) as the best example of this colour, not any of the large range of other blue and green shades to which the term can refer. A similar situation exists with other BCTs: their boundaries vary widely, but their focal hues show clear cross-linguistic convergence. This is a remarkable finding. It suggests that colour terminology is not an unconstrained free for-all. Basic colour terms in all languages target a restricted range of colours.
The restriction of focal hues to a small set was not Berlin and Kay’s only discovery. They also found that the number of BCTs in a language makes it possible to predict exactly what the basic colour terms would be. For example, consider those languages with only two colour terms. Berlin and Kay claimed that such languages have one term for white, red and yellow, and another for black, green and blue. No languages with only two BCTs have any different combinations: for example, no language has as its two BCTs one referring to white, green and blue, and the other to red, yellow and black. Similar generalizations were also made for larger BCT systems as well. In general, Berlin and Kay’s findings were that the range of colour term references in the languages of the world represents a tiny selection out of the myriad theoretically possible options. Languages don’t just choose their colour terms arbitrarily, but observe quite strict constraints in the range of possible focal colours to which their BCTs refer. Berlin and Kay interpreted these findings as an evolutionary sequence. Colour vocabularies started with just two terms, and developed diachronically into more ramified, complex systems, with different hues splitting off to create independent colour categories. Table 11.1 sets out these findings, showing the five different evolutionary ‘stages’ into which colour-naming systems can be arranged. The colour names (given in obvious abbreviations) refer to the best examples of the colours.
Terms for brown, purple, pink, orange and grey are added in two later stages, giving the maximal system of eleven colour terms observed cross linguistically, arranged in a seven-stage system.
It is important to be clear about what these stages mean. First, note that it is not the case that a language cannot refer to any colour incompatible with its typological stage. Just because a language belongs to stage V, for example, does not mean that speakers do not acknowledge brown, orange, purple or pink hues as genuine colours. All it means is that these colours are treated as variants or types of one of the named colours in the language, and thus fail to count as basic colour terms. Second, note that the focal colours for some of these categories vary by both language and speaker. We can illustrate this with the different focal colours for the Green/Blue category found at Stages III and IV. Speakers of some languages choose blue as the focus, speakers of some choose green, speakers of some choose both, and speakers of others chose either one, but not both (MacLaury 1999: 5). Categories like this with more than one focal colour are called composite categories.
We don’t have to go along with Berlin and Kay’s evolutionary interpretation of their own findings. The five stages can be interpreted simply as typological generalizations stating the references of colour term systems of different sizes. Here are some examples of the stages (from Kay et al. 1997):
The seven-stage typology revealed by Berlin and Kay’s research has been broadly confirmed (MacLaury 1999: 30). This doesn’t mean, however, that it’s always easy to tell what stage of colour vocabulary a language instantiates. Since languages are in a continual state of change, there will often be transitional cases which complicate the analysis. For example, a language will accomplish the transition from one colour stage to another by introducing a new, special term which only gradually becomes a BCT, and it may well be hard to decide exactly when the transition is complete. But this is no more than a typical problem encountered in any attempt to distinguish typologically significant generalizations in the flux of language variation and change. On the other hand, there are some significant counterexamples to Berlin and Kay’s typology, as well as fundamental criticisms of their methodology. We will explore each in turn.
As acknowledged by Kay and Regier (2003: 9085), some exceptions to the Berlin and Kay findings have come to light, and the original typology cannot any longer be claimed as universal. Some of the exceptions necessitate only minor adjustments. Thus, Russian (Indo-European; Russia) has 12 BCTs, one more than the maximal number originally recognized, including goluboj ‘light, pale blue’ and sinij ‘dark, bright blue.’ Hungarian (Finno-Ugric; Hungary) has both piros ‘light red’ and vörös ‘dark red’ BCTs (MacLaury 2002: 499), not a possibility accommodated in the original system.
More seriously, the Salishan languages of the Pacific North West of North America include a yellow-with-green BCT, which does not fi t any of the predicted types (MacLaury 1999: 20–21). Further, parameters not considered in Berlin and Kay’s original investigation, such as brightness, seem to form the basis of basic colour terms in some languages (see MacLaury 1999 for discussion), a possibility which challenges the original decision to exclude this parameter from the investigation. Another serious challenge to Berlin and Kay’s findings comes from investigation of Yélî Dnye (isolate; Papua-New Guinea). This language appears not to have any colour terms which would count as basic on Berlin and Kay’s criteria. This is because Yélî Dnye colour terms are often simply the reduplicated names of objects; for instance, the word glossable as ‘red’, mtyemtye, comes from mtye, ‘red parrot species’, and kpêdêkpêdê, glossable as ‘black’, is a reduplication of kpêdê ‘tree species’. Furthermore, large zones of the Munsell colour space are simply unnamed, lacking any distinct term (Levinson 2001). This poses a major challenge to the premises of the Berlin and Kay investigation.
Counterexamples like those above have led to the Berlin and Kay typology being restated not as a universal of colour semantics, but as merely a particularly strong cross-linguistic tendency, to any part of which exceptions will exist. This does not remove its value: there are very few, if any, areas in language where iron-clad generalizations are possible. A more serious type of challenge is one which questions the very basis of Berlin and Kay’s colour survey. A number of such challenges have been made. For example, MacLaury (1999: 19) notes that Berlin and Kay’s findings may have been skewed by the nature of the Munsell colour system itself, which does not allow representation of a psycho logically important dimension of colour perception, luminosity. Luminosity is non-reflective brightness originating within the source of the colour itself, for example the sun, or a hot, glowing object. It contrasts with lightness, which is reflected illumination. Since the Munsell chips are only refl ective, ‘they may not adequately reveal the meanings of certain color terms that principally name luminosity’ or involve luminosity as a crucial factor. Similarly, Lucy (1997) criticizes the Berlin and Kay tradition on the grounds that it usually omits consideration of the colour term’s characteristic referential range, simply assuming that the BCTs elicited are in fact primarily used for the coding of colour (see Lucy 1997: 322–333). Lucy claims that this ignores the way ‘colour’ terms are actually used in a language. The set of Munsell chips used in the study is not at all representative of the everyday contexts in which colour vocabulary is used. We rarely use colour terms in the context of an abstract exercise in hue-naming. Typically, a colour term will be predicated of a real object, which necessarily introduces many other considerations as possible determinants of its use. But all these other components are factored out by the use of the Munsell array. ‘In a sense’, Lucy says, use of the set of Munsell chips dictated in advance the possible meanings the terms could have since no other meanings were embodied in the samples. Although restricted in this way, the stimulus array was also very complex, and the labeling task performed with it forced informants to make referential microcomparisons and judgements of a sort rarely encountered in daily life. The task assumed that speech is about labeling accuracy rather than situational intelligibility . . . (Lucy 1997: 333)
Lucy’s criticism, in other words, is that Berlin and Kay’s methodology artificially constituted ‘colour’ as a meaningful category in the languages they tested, without any attention to the contextual values of colour terms, or consideration of non-colour uses – uses which might, after all, be extremely revealing about the semantics of the terms.
QUESTION Does this seem a fair criticism? Why (not)?
One obvious example of the selectivity of the Berlin and Kay scheme is discussed by Payne (2006: 605). Some languages’ colour words also include reference to factors which fall outside the domain of colour pure and simple. For example, in Maasai (Nilo-Saharan, Kenya/Tanzania), there are colour-plus-design terms for ‘spotted black and white’, ‘thinly striped, typically with tan and white’. These would presumably not count as basic colour terms for Berlin and Kay, but this raises exactly the point in question: how far does ‘colour’ reflect a psychologically or culturally real category?
This point can be most clearly seen in work done by Conklin (1964) on colour terms in Hanunóo (Austronesian; Philippines). In Berlin and Kay’s terms, Hanunóo has a stage III colour system, with categories translated ‘black’, ‘red’, ‘white’ and ‘light green’. But Conklin’s account shows that translation simply with English colour adjectives makes the wrong predictions about what the terms will be used to refer to, since the words in question have other semantic values which it is crucial to take into account. In fact, Conklin shows, Hanunóo ‘colour’ terms refer to three other parameters as well as hue: a light/dark opposition; a dryness/wetness (freshness) one, and a deep versus pale distinction. The reason that these other values matter is that they are just as important as the hue dimensions in governing what Hanunóo colour terms refer to. For example, a shiny section of newly cut bamboo, which English speakers would describe as brown, is described in Hanunóo as malatuy ‘green’ (Conklin (1964 [1955]: 191). This is extremely surprising, if we assume that malatuy and related terms have hue as their basic reference. But if we include the three other dimensions in our description of the terms’ meanings, we have a way of understanding what is going on. Instead of malatuy meaning ‘green’, it really means something like ‘wetness’, a description which explains its application to the newly cut bamboo – and also to many green things as well. As Lucy puts it (1997: 326), ‘[w]hat is crucial to recognize here is that an “adequate knowledge” of the system would never have been produced by restricting the stimuli to color chips and the task to labeling’ (italics original). Berlin and Kay’s colour elicitation methodology simply presupposes that words which can be used to refer to Munsell chip categories are basically colour terms; Conklin’s research suggests that this may seriously misrepresent the semantics of an individual language. Only investigation of actual, natural discourse will reveal the semantics of ‘colour’ terms in a language, and any attempt to erect universals of colour vocabulary that does not include such investigation risks prejudging the data.
