Printer Friendly

A study of colour grouping in three languages: a test of the linguistic relativity hypothesis.

The linguistic relativity hypothesis (LRH) (Sapir, 1921; Whorf, 1956) is currently out of favour. For instance, modern textbooks mention it as an interesting episode in the history of psychology, but then generally dismiss it (see, for example, Atkinson, Atkinson, Smith, Bern & Hilgard, 1990, p. 327).(1) This current status of the LRH is due largely to the influence of just two studies, both concerned with colour: Heider's (1972) study of the Dani of New Guinea and Berlin & Kay's (1969) survey of the colour lexicons of 98 languages. These two studies were seminal and timely, and they have become citation classics. Nevertheless, we shall argue that despite the evidence from these studies, the case against the LRH is weak, and that the thesis deserves further empirical tests. We then report such a test that compares speakers of English, Russian and Setswana using a colour grouping task.(2)

Until about 1970, it was taken as virtually axiomatic that speakers of languages with different inventories of colour terms had different cognitive representations of colour: linguistic differences implied parallel differences in non-linguistic aspects of colour cognition (see, for instance, Brown & Lenneburg, 1954). Since about 1970, there has been a shift in the Zeitgeist away from axiomatic acceptance of the LRH towards belief in colour universals (Brown, 1976). A major influence on this shift was Heider's (1972) study of the Dani of New Guinea. She found that even though the Dani had only two basic colour terms - mill 'dark' and mola 'light' - they remembered focal exemplars (the best examples) of eight English colour terms better than non-focal exemplars. Further, they learned names for the focal exemplars more effectively than for the non-focal exemplars. Heider interpreted her results as reflecting the universal perceptual distinctiveness of the focal exemplars, perhaps derived from universal perceptual physiology (De Valois & Jacobs, 1968).(3) However, although the results are consistent with the universalists' position, they can also be interpreted as supporting the LRH. Although the Dani remembered focal colours better than non-focal colours - consistent with universalism - their performance was much worse than the American comparison group, which could reflect the Dani's limited repertoire of basic colour terms, consistent with the LRH (see Ratner, 1989).(4)

A second major influence on the shift in the Zeitgeist was Berlin & Kay's (1969) monograph. Before 1969, the marked differences in the number of colour terms, and in the denotations of colour terms across languages, were interpreted as strong evidence for the cultural relativity of colour categories. It was assumed that the continuum of the colour spectrum was segmented into colour categories 'without constraint' (Gleason, 1961). Berlin & Kay investigated the colour terms of 98 languages by asking participants to 'map' the denotations of colour terms in a Munsell array of 330 colours, and to select the best example (the focus) of each colour term. Berlin & Kay found that while there was considerable variation across languages in which the boundaries of colour categories occurred, there were just 11 delimited regions of colour space where the foci of colour categories were to be found. They interpreted their results as showing that the 'basic colour terms' (most frequent, salient and high consensus) of all the 98 languages were drawn from a universal inventory of just 11 colour categories: BLACK, WHITE, RED, GREEN, YELLOW, BLUE, BROWN, PURPLE, PINK, ORANGE, GREY. The crucial invariant property of these categories is their focus (the best exemplar), rather than their boundary. (These hypothetical universal categories are denoted here in small capitals by the English term used for the 11 focal regions of colour space). Kay & McDaniel (1978) argued that the 11 universal categories are based on universal perceptual physiology (see also Heider, 1972): the perceptual structure of colour space varies less across cultures and languages than does the linguistic expression of colour. Thus, if the perceptual structure of colour space is broadly universal, speakers of languages with differing numbers of colour terms should show strong similarities in how they sort colours into similarity groups. Further, speakers of languages with less than 11 basic colour terms should have 'perceptual categories' corresponding to the missing universal categories. In a sense, there are nascent linguistic categories ready to form at physiologically determined 'faults' in colour space. Thus, for instance, if speakers of a language with no separate terms for GREEN or BLUE (such as Setswana, which we include here) were asked to sort colours in the GREEN - BLUE region into two groups based on their perceptual similarity, they should tend to sort them into one group focused around GREEN and one group focused around BLUE, just as speakers of languages such as English or Russian that do have separate GREEN and BLUE terms would do.

However, even if there are colour universals based on common physiology, this does not exclude the possibility that colour cognition might also be modulated by language. Sensitivity to basic visual dimensions can be improved with practice (Fahle & Edelman, 1993), and learning to categorize a perceptual dimension, such as brightness, improves sensitivity to that dimension (Goldstone, 1994). Further, there is evidence supporting the case for the 'categorical' perception of colour (see Harnad, 1987): within-category discriminations are worse than equivalent between-category discriminations (Bornstein & Korda, 1984, 1987; Boynton, Fargo, Olson & Smallman, 1989; Kay & Kempton, 1984). Language learning could provide the engine for perceptual learning in the colour domain. In order to use colour terms with competence, the learner must attend to the key attributes of a term: the focus and the boundary. Thus if colour perception is changeable, sensitivity to these key regions is likely to increase. Further, to the extent that languages differ in the number of basic colour categories, or in the position of category boundaries, there should be differing distributions of sensitivity across colour space reflecting the differing distribution of significant category attributes.

Davies & Corbett's (1997) data are consistent with language modulating universal colour cognition. We compared speakers of English, Russian and Setswana on a colour sorting task. The three languages differ in their number of basic terms (English has 11, Russian has 12 and Setswana has five) and in the position of some of their category boundaries.(5) Our informants were asked to sort 65 coloured tiles into groups, so that members of a group resembled each other, 'in the same way that members of a family resemble each other'. The dominant pattern in the results was that colour sorting was broadly similar across the three languages. However, there were also significant differences between the language samples that were consistent with linguistic effects. The Batswana(5a) (who have a BLUE or GREEN term, botala) were more likely to group BLUE with GREEN, than either of the other two groups. In addition there were a series of small, but significant, differences that related to the difference in positions of category boundaries across languages. The results were thus consistent with weak linguistic relativity.

We report here the second stage of the colour grouping task summarized above. Participants were again asked to sort the tiles into groups, but this time, the number of groups (N) was specified; N varied from 2 to 12. If there are linguistic effects on colour grouping, then there should be an association between how tiles are named and how tiles are grouped. Such an association would result in differing patterns of colour grouping across the three languages. Conversely, to the extent that colour perception is universal there should be clear similarities across languages in how the colours are grouped. We explore these general questions in three specific ways. First, we test whether levels of consensus within a language group over which colours to group together varies with N. If grouping is influenced by colours sharing a common name, then the level of consensus over which tiles to group together should fall once N exceeds the number of basic colour terms in the language. Thus the level of consensus for Setswana should start to fall when N is greater than five. In contrast, no such fall should happen for English or Russian; rather, if anything, if grouping is influenced by colour naming, levels of consensus should peak at N = 11 for English and N = 12 for Russian.

Second, if there are linguistic effects on colour grouping, there should be an association between tile-similarity matrices derived from how participants group colours, with tile-similarity matrices derived from how they named the tiles. The intra-language correlations between the grouping and the nominal matrices should be higher than any interlanguage correlation between grouping matrices and nominal matrices. On the other hand, if colour cognition is universal, then the groups of colours formed by speakers of the three language groups should be broadly similar, and there should be strong correlations between grouping-similarity matrices across languages.

Third, if there is a linguistic influence on colour cognition, then the groups formed should differ across the three languages reflecting the differences in the positions of colour category boundaries. In looking for such effects we concentrate on the BLUE - GREEN region because there is a clear three-way difference in how the languages categorize these colours: English divides the region into two categories - blue and green; Russian divides the region into three categories - zelenyj 'green', goluboj 'light blue' and sinij 'dark blue'; while Setswana has a single term for the region - botala 'grue 'blue or green'. If there are linguistic influences on grouping, then the Batswana (Setswana speakers) should be the most likely to group BLUE tiles with GREEN tiles, while the Russians should be most likely to form separate DARK BLUE and LIGHT BLUE groups.



There were three samples of participants: an English-speaking sample from Britain; a Russian-speaking sample from Russia; and a Setswana-speaking sample from Botswana in southern Africa. In all cases, the participants were mother-tongue speakers of the appropriate language and none was fluent in a second language. The samples are designated here by the mother-tongue spoken by each sample.

There were 47 people in the English sample, 24 men and 23 women, and their ages ranged from 21 to 65 years with a mean of 29 years. They lived in and around Guildford in the south of Britain. There were 77 people in the Russian sample, 24 men and 53 women and their ages ranged from 18 to 65 years with a mean of 34 years; they lived in Moscow. There were 44 people in the Setswana sample, half were men, and half were women, and their ages ranged from 26 to 82 years, with a mean of 45 years. They lived in villages around the town of Kanye in the south of the country, and they were from the Bangwaketse tribe.


The stimuli consisted of 65 coloured tiles displayed against a uniform grey background. The tiles were 50 mm squares of coloured paper sprayed with a light film of transparent varnish, and mounted on a thin rigid base. The colours were an evenly spread sample of 'colour space' taken from the Color-Aid Corporation range. Figure 1 shows their locations in the CIE 1976 uniform chromaticity diagram (u', v') together with the loci of the 11 'universal' foci that may be used as landmarks to interpret the space? (The full technical specification of the stimuli is available in the Appendix to Davies & Corbett, 1997). The illumination was not controlled rigidly, but the experimenters avoided deep shade and direct sunlight, and they checked that the illumination was sufficient for them to pass the City University Colour Vision Test without effort (the test requires illuminant c).

Language and instructions

The experiment was conducted in the appropriate mother tongue for each sample by a native speaker of the appropriate language in most cases.(7) The English instructions were translated into the other languages by a native speaker of that language, and back-translated into English by another native speaker and the cycle repeated until the match was satisfactory.


All participants did five tasks: a colour term 'list task', the City Colour Vision Test (Fletcher, 1980), a ' free-sorting' task, a ' forced-sorting' task, and a colour naming task, in that order. The focus of this paper is on performance on the forced-sorting task, and its relationship to colour naming; the other tasks have been reported elsewhere (Davies et al., 1992; Davies & Corbett, 1994, 1995, 1997; Davies, Laws, Corbett & Jerrett, 1998).

For the forced-sorting task the tiles were placed in random order on a tray covered in grey cloth and participants were asked to sort the tiles into 'N' groups, so that 'members of a group looked similar to each other, just as members of a family resemble each other'. Participants sorted the tiles four times in this way differing only in the value for N. The value of N ranged from 2 to 12 with each value being used about the same number of times within a language group. The values for N for each participant were chosen so that they were more or less equally spaced, within the overall constraint that the values of N should be used as equally often as possible. Thus typical sets of values for N were: 3, 6, 9 and 12; and 4, 7, 10 and 12. Participants were randomly assigned to a set of four values for N, and the order they did their four sortings was also randomized. These assignment procedures meant that the numbers of participants for each value of N for each sample were: either 17 or 18 for the English group; 28 for the Russian group; and 16 for the Setswana group.


Similarity matrices

The analysis was based on two kinds of similarity matrices: 'grouping matrices' and 'nominal matrices'. Grouping matrices were derived from the participants' performance on the grouping tasks. The cell entries in the matrices are the number of those who placed a given pair of tiles in the same group, for each possible pair of tiles ((65 x 64)/2 pairs). A grouping matrix was constructed for each number of groups (N), for each language, resulting in 11 such matrices for each language. The nominal matrices were derived from the tile naming data, and the similarity scores were the extent to which a given pair of tiles were given the same name. The number of participants that used each colour term for each tile (the frequency) was used to derive the similarity score. Specifically, the similarity score was the sum of the smaller of the two frequencies across the two tiles for each instance of a common name being used for those tiles. For example, if tile A was called red by 20 people and pink by 5 people, and tile B was called red by 17 people and pink by 7 people, then the nominal similarity score would be 17 (red)+ 5 (pink) = 23.

Levels of consensus in grouping

The standard deviation ([Sigma]) of the grouping matrices is a measure of the intra-sample consistency in tile grouping. The closer to unanimity over which tiles to group together (proportional scores tend towards 1 or 0) the higher [Sigma] will be. The values of [Sigma] for each language and for each level of N are shown in Fig. 2 together with the maximum possible values and the expected values if grouping was done randomly.(8)

The functions for the three languages are well above the random function: all variance ratios for each language compared to the corresponding random variance for each level of N are statistically significant (F(2080, 2080) [greater than] 4.0, p [less than] .01). The functions for English and Russian have minima at N = 2, and rise to maxima at N = 5 and N = 4 respectively, and then decline more or less monotonically, through to N = 12. In contrast, the Setswana function has a maximum at N = 2, and then declines more or less consistently through to N = 12. The reliability of the differences that can be seen in Fig. 2 can be scaled in terms of the variance ratios that are statistically significant: ratios of [greater than] 1.4 are significant at the .05 level and of [greater than] 1.6 are significant at the .01 level. Thus the difference between the Setswana score for N = 2 and both the English and Russian scores at N = 2 can be seen to be significant. In addition, each of the variance ratios comparing English and Setswana from N = 5 to N = 12 is also significant.

Similarity matrix structure

1. Correlations between grouping matrices. The degree of correspondence between the grouping matrices was assessed using the correlation coefficients between sets of corresponding cells (2080 cells) in pairs of matrices. These correlation coefficients were calculated for each pair of languages, for each level of N (11 x 3 correlations). The mean inter-language correlation across N was r = .803 (range .685 to .918). (These, and all of the correlations reported below, are statistically significant at the .01 level at least.) Although these correlations are relatively high, they were generally lower than the intra-language correlations for successive values of N(N = 2 with N = 3, N = 3 with N = 4 etc). The mean intra-language correlation was .947 for English, .833 for Russian and .890 for Setswana. Thus, although there is common structure in the grouping matrices across languages, there is some additional intra-language influence, beyond the common interlanguage effects.

2. Correlations between nominal and grouping matrices. Figure 3 shows the correlation coefficients between the grouping matrices for each level of N, and the appropriate nominal matrix, for each language. The English and Russian correlations are higher than the equivalent correlations for Setswana for each level of N (significant at the .01 level at least), and these differences increase more or less consistently with N; the Setswana function tends to level off after N = 7, whereas the functions for the other two languages continue to rise. The Russian function tends to lie above the English function, except for N = 2 and the differences at N = 2, 5, 7, 9, 11 and 12 are significant at the .05 level at least. In general, the nominal grouping correlations are lower than the equivalent intra-language correlations for adjacent values of N. The difference between these two sets of correlations averages 0.308 for English, 0.171 for Russian and 0.420 for Setswana.

Cluster analysis of groups

Figures 4-6 show the results of cluster analysis of the grouping matrices for each level of N for English, Russian and Setswana, respectively. The cluster analyses partition the 65 tiles into N clusters, and the cluster membership is shown plotted in CIE 1976 (u[prime], v[prime]) space. Figure 1 shows the loci of the tiles and the universal foci in the same coordinates as Fig. 4, and the location of the universal foci can be used as landmarks to interpret the content of the clusters. We have tried to use the same symbols to denote the same regions of colour space across the various figures. For example, Fig. 4a shows two clusters which partition the space approximately into low values of u[prime] (+) and high values of u[prime] (x). Very similar results can be seen for Russian [ILLUSTRATION FOR FIGURE 5A OMITTED] and for Setswana [ILLUSTRATION FOR FIGURE 6A OMITTED], and the same symbols are used to denote corresponding regions in the three graphs.

There are two general points of note. First, although the full CIE colour space is three-dimensional, in most cases, the clusters form non-overlapping regions in the two-dimensional space (u[prime]v[prime]) without recourse to the lightness dimension ([L.sup.*]). Second, there is striking similarity across languages for the same levels of N. This similarity is most marked for low values of N, but it persists right through to 12 clusters. This impression is supported by formal measures of association between cluster membership, at each level of N, for each pair of languages. The mean and (range in brackets) of Cramer's V across levels of N were: English-Russian 0.875 (0.767 to 0.976); English-Setswana 0.761 (0.625 to 0.940); Russian-Setswana 0.777 (0.679 to 0.844); p [less than] .000009 in all cases.

Boundary effects

1. Grouping of GREEN and BLUE. The universal foci (except PINK, BLACK, WHITE and

GREY) tend to be the most saturated instance of a category; thus they will tend to lie towards the periphery of the u[prime], v[prime] region formed by the 65 tiles. For instance, the universal GREEN fOCUS (0.11, 0.49) has lower u[prime] values than any of the 65 tiles, but its neighbours are also included in GREEN. Similarly, the universal BLUE focus (0.15, 0.35) has lower u[prime] than its immediate neighbours, but they are also included in BLUE. If we operationally define BLUE and GREEN as the regions adjacent to the respective universal foci, then it can be seen from Figs 4-6 that separate BLUE and GREEN clusters form later for Setswana than for the other languages. For Setswana [ILLUSTRATION FOR FIGURE 6 OMITTED] GREEN and BLUE are in the same cluster from N = 2 to N = 6 (designated by x); they then separate at N = 7 (+, O) but rejoin at N = 8. For Russian and English, GREEN and BLUE separate at N = 4 (+, O) and distinct BLUE and GREEN clusters are present from N = 5 to N = 12. Thus it appears that Setswana is more likely than English or Russian to group BLUE and GREEN tiles together from about N = 5 to N = 8. In order to test this impression more formally, inter-BLUE - GREEN scores were calculated for each participant, for each level of N. These scores were the number of times each possible pair of tiles across the two 'core' regions were grouped with each other, expressed as a proportion of the maximum possible score.(9) Figure 7 shows the mean inter-BLUE - GREEN scores across participants for each level of N and for each language.

The inter-BLUE - GREEN scores decline sharply with N, approaching zero for high values of N. Further, for N = 2 through to N = 8 it appears that the inter-BLUE - GREEN scores tend to be larger for Setswana than for English or Russian. ANOVA on language (3) by N (11) showed that both main effects and the language by N interaction were significant: language (F(2,623) = 36.8; p [less than] .0009); N(F(10,623) = 26.9; p [less than] .0009); interaction (F(20,623) = 1.81; p [less than] .02). Comparisons of the three languages on the inter-BLUE - GREEN scores for each level of N found that the differences approached significance for N = 2, 3 and 4 (p between .06 and .11); there were significant differences for N = 5, 6 and 8 (F(2,56-60) [greater than] 8; p [less than] .002); there were no significant differences for N = 7 (p = .07), or from N = 9 onwards. For the cases where there were significant differences, the English and Russian scores were significantly less than Setswana scores at the .05 level (Tukey's HSD).

2. Grouping of DARK BLUE and LIGHT BLUE. From Figs 4-6, it can be seen that the BLUE region (designated by O) tends to remain as a single coherent region once it has separated from the GREEN region. However, for English at N = 12, the BLUE region splits into two categories which correspond broadly to LIGHT BLUE (O) and DARK BLUE ([symmetry]). The DARK BLUE category includes two tiles from the BLUE group at N = 11 plus one of the purple tiles, and this category has low scores on L* the lightness dimension relative to the BLUE tiles. In contrast, there is no suggestion of Russian forming two BLUE categories at any stage, and the same applies to Setswana.(10)


The most striking pattern in the results is the similarity in the behaviour of the three language samples. This is shown by the high inter-language correlations between the grouping similarity matrices and by common patterns of subdivision of colour space found in the cluster analyses [ILLUSTRATION FOR FIGURES 4-6 OMITTED]. In addition to this broad interlanguage correspondence, there are also some notable differences in the behaviour of the three samples.

First, the languages differed in the level of consensus over which tiles to group together. English showed more consensus than Russian which in turn showed more consensus than Setswana. The low consensus score for Setswana is consistent with the relatively low salience of the concept of colour in Batswana culture. On the other hand, it is not apparent why there should be a difference in consensus between English and Russian.

The relationship between the level of consensus and the number of groups also differed for the three languages. The clearest difference is that Setswana had maximum consensus at JU = 2, whereas the other two languages had their lowest levels of consensus at N = 2 [ILLUSTRATION FOR FIGURE 2 OMITTED]. English and Russian had their maximum consensus scores at N = 4 and N = 5 respectively, and the level of consensus - particularly for English - remained relatively high through to N = 10. Further, the correlations between the nominal similarity matrices and the grouping similarity matrices peaked at higher values of N for English and Russian than for Setswana (N = 10, 12, 7, for English, Russian and Setswana respectively; [ILLUSTRATION FOR FIGURE 3 OMITTED]). This is probably a consequence of Setswana having just five basic colour terms whereas they were required to sort the tiles into more than five groups (up to 12). Once N is greater than five, it is increasingly likely that tiles that share a name will be placed in separate groups, and thus the correlation should fall. However, the peak correlation occurs for seven groups rather than five groups. Although there are just five basic colour terms in Setswana, there are a number of 'secondary' colour terms, some of which are close to being basic. The three most common of these additional terms are' sethunya 'yellow-blossom', seaole 'purple' and bosethla 'light-brown'. Thus there is some remaining scope for the association between colour grouping and nominal grouping to increase beyond five groups. The correlations for English and Russian peak at higher values of N than for Setswana - at 10 and 12 respectively - which is consistent with the greater number of basic colour terms in the two languages being greater than those in Setswana, and with colour grouping mirroring lexical category membership. In addition, the size of the nominal grouping correlations were consistently greater for English and Russian than for Setswana. This may simply reflect the lower level of consensus over grouping for Setswana, but it is also consistent with the low salience of the concept of colour in Setswana.

There are also predicted differences in the likelihood of grouping BLUE and GREEN tiles together. Setswana is more likely to group BLUE tiles with GREEN tiles than the other two languages. BLUE and GREEN form separate clusters by N = 5 for English and Russian, but a conjoint BLUE - GREEN cluster persists through to N = 8 for Setswana. This impression was supported by the analyses of variance which showed that the inter-BLUE - GREEN scores were higher for N = 5 to N = 8 for Setswana than for English or Russian [ILLUSTRATION FOR FIGURE 7 OMITTED]. On the other hand, there was no support for the prediction that Russian should have separate DARK BLUE and LIGHT BLUE clusters, reflecting the language's two basic terms for the BLUE region.

We believe that we have established that there are reliable differences between the languages in how the tiles are grouped. The size of these effects is often small relative to the overall similarity between language groups, but they are associated with differences in the colour category structures between the languages. The data supports the modest linguistic relativity of Kay & Kempton (1984). However, it is unclear what the locus (or loci) of the effect is. Perhaps the least interesting possibility is that respondents use a direct language strategy to group the tiles. If this was so, then language would be affecting behaviour, but strictly speaking it may not be affecting cognition. A second possibility is that colour categorization is affected by the different linguistic structures. Informants' propensity to group perceptually different colours together might be structured by language. A third possibility is that, as we speculated in the introduction, learning the language has in some sense changed some aspect of colour perception. Consistent with the latter possibility, our respondents almost always said they used a strategy based on colour appearances. However, the marked similarity in grouping among the three language samples for high levels of N in the cluster analysis [ILLUSTRATION FOR FIGURE 4, 5, AND 6 OMITTED] and in BLUE - GREEN similarity scores [ILLUSTRATION FOR FIGURE 7 OMITTED] is consistent with equivalent perceptual structures across language samples. When the Batswana have to produce more groups than they have readily available linguistic labels, their groups look much like the English and Russian groups. This similarity probably derives from similar perception. Thus perhaps the safest conclusion is that the data support linguistic influences on grouping, but the case is not proven for perception.

However, drawing a distinction between perception and cognition is notoriously difficult (some would say meaningless). To constructionists in particular, perception is heavily influenced by top-down conceptual processes. In some respects, constructionists and linguistic relativists are natural allies, and to them both, the issue of the locus of the effect may be unimportant. On the other hand, we can think of tasks as varying in their degree of conceptual influence. Sensation, in the traditional sense, is less influenced by conceptual processes than perception. Similarly, the relative task demands on memory can be varied. We are carrying out a series of experiments that vary the relative perceptual and memory loads in order to map the nature of linguistic effects more precisely, and to perhaps provide a handle on the perception versus cognition question.

In summary, there are differences in colour categorization associated with differences in the language. Setswana was less likely to group BLUE - GREEN tiles than the other two languages. On the other hand, there was no support for the prediction that Russian should form separate DARK BLUE (sinij) and LIGHT BLUE (goluboj) groups, and what differences there were were small compared to the broad similarity among the three samples' behaviour. The data are consistent with strong universalism modulated weakly by language.


This work was supported by the ESRC Grant R000 23 1958 which is gratefully acknowledged. I am indebted to our collaborators in Botswana - David and Tiny Jerrett - and in Russia - Vladimir Moss and Svetlana Stepanskaja. Catriona MacDermid helped to organize the data collection and the initial data analysis, and I am grateful for her contribution. My colleague Greville Corbett supervised the project with me and, as ever, his advice, particularly on linguistic matters, has been invaluable. I would also like to thank two reviewers for their speedy and constructive comments.

2 The study we report here is part of a programme of research addressing the linguistic relativity question. Two recent papers (Davies & Corbett, 1997 and Davies, Sowden, Jerrett, Jerrett & Corbett, 1997) review the key issues in the literature, whereas here we focus on issues of direct relevance to the current experiment.

3 See De Valois & De Valois (1993) for a current version of this model.

4 Lucy & Schweder (1979) also question the validity of Heider's main universalist findings on methodological grounds; but see also Garro (1986).

5 English has 11 basic colour terms: white, black, red, green, yellow, blue, brown, purple, pink, orange and grey (Davies & Corbett, 1995). Russian has 12 basic colour terms: belyj 'white', cernyj 'black', krasnyj 'red', zelenyj 'green', zeltyj 'yellow', sinij 'dark blue ', goluboj' light blue ', koricnevyj 'brown ', fioletovyj 'purple', rozovyj 'pink', oranzevyj 'orange', seryj 'grey' (Davies & Corbett, 1994). Setswana has five basic terms: bosweu 'white', bontsho 'black', bohibidu 'red', botala 'green or blue' and borokwa 'brown' (Davies et al., 1992).

5a The country is Botswana, the language Setswana, and the people are the Batswana (Singular (Motswana)).

6 Within the CIE system the total colour is made up from red, green and blue components, and the proportions of these three must sum to one. The CIE chromaticity coordinates can thus be thought of as the proportions of red (x), and green (y), in each colour; a third coordinate, luminance or reflectance (Y) makes up the CIE tri-stimulus values. By implication, the proportion of blue light is given by 1 - (x + y). Every possible colour has a unique locus in three-dimensional (Y, x, y) space and these tri-stimulus values may be used to convert the Color-Aid stimuli into the more familiar Munsell, or OSA systems, through conversion tables in, for instance, Newhall, Nickerson & Judd (1943). Note, however, that this tri-stimulus space is not a perceptually equal space; that is, equal distances do not correspond to equal perceptual steps. The CIE 1976 system (u[prime], v[prime]) represents colours in a transformed space which is approximately perceptually equal. In the 1976 space u' is a transformation of x, and v[prime] is a transformation of y. For instance, in Fig. 1, the universal blue has coordinates of (u[prime] = 0.18, v[prime] = 0.19). The proportion of blue is thus: 1 -(0.18 + 0.19) = 0.63. Thus, as would be expected, the universal blue has a high proportion of blue in it, and blue colours are to be found towards the origin of the graph (low u[prime] [red] and v[prime] [green]). On the other hand, red colours have high proportions of red in them (u[prime]) and are to be found towards the right of the graph. The positions of the 11 universal foci in Fig. 1 can be used to interpret the remaining regions of the CIE chromaticity diagram. (See Hunt, 1987, for further information on the CIE system.)

7 For Russian, most participants were tested by a native speaker, but some were tested by a fluent second-language speaker. The essential pattern of results did not differ across the two experimenters.

8 The maximum possible value of [Sigma] varies with N and with how equal groups sizes are. There were only small differences between the language samples in the distribution of group sizes, and the group sizes tended to be equal; we therefore give the maximum possible value of [Sigma] assuming group sizes were as equal as possible. In fact, the argument we make still stands even if we take variation in group size into account.

9 The core GREEN region consists of those tiles which were labelled with GREEN terms (green, or zelenyj or botala) by half or more of at least two of the language groups. There were five tiles that met these criteria: GBG-hue, G-hue, GYG-hue, YG-hue and YGY-hue. The equivalent criteria for the BLUE region produces four core BLUE tiles: B-hue, BGB-hue, B-T1 and BGB-T3. The inter-BLUE - GREEN region is then the cross-product of the two sets of tiles just given, consisting of 20 cells.

10 This assertion is supported by the equivalent ANOVA to that reported for the inter-BLUE-GREEN region reported earlier; there were no significant effects involving the language factor.


Atkinson, R. L., Atkinson, R. C., Smith, E. S., Bern, D. J. & Hilgard, E. R. (1990). Introduction to Psychology, 10th ed. New York: Harcourt Brace Jovanovich.

Berlin, B. & Kay, P. (1969). Basic Color Terms: Their Universality and Evolution. Berkeley and Los Angeles: University of California Press.

Bornstein, M. H. & Korda, N. O. (1984). Discrimination and matching within and between hues measured by reaction times: Some implications for categorical perception and levels of information processing. Psychological Research, 46, 207-222.

Bornstein, M.H. & Korda, N. O. (1987). Some psychological parallels between categorisation processes in vision and audition. In S. Harnad (Ed.), Categorical Perception. New York: Cambridge University Press.

Boynton, R. M., Fargo, L., Olson, C. X. & Smallman, H. S. (1989). Category effects in color memory. Color Research and Application, 14(5), 229-234.

Brown, R. (1976). In memorial tribute to Eric Lenneberg. Cognition, 4, 125-153.

Brown, R. & Lenneberg, E. H. (1954). A study in language and cognition. Journal of Abnormal and Social Psychology, 49, 454-462.

Corbett, G.G. & Jerrett, D. (1996). Cross-cultural differences in colour vision: Acquired colour blindness in Africa.

Davies, I. R. L. & Corbett, G. G. (1994). The basic color terms of Russian. Linguistics, 32, 65-89.

Davies, I. R. L. & Corbert, G. G. (1995). A practical field method for identifying probable basic colour terms. Languages of the World, 9(1), 25-36.

Davies, I. R. L. & Corbett, G. G. (1997). A cross-cultural study of colour grouping: Evidence for weak linguistic relativity. British Journal of Psychology, 88, 493-517.

Davies, I. R. L., MacDermid, C., Corbett, G. G., McGurk, H., Jerrett, D., Jerrett, T. & Sowden, P. (1992). Color terms in Setswana: A linguistic and perceptual approach. Linguistics, 30, 1065-1103.

Davies, I. R. L., Sowden, P., Jerrett, D. J., Jerrett, T. & Corbett, G. G. (1998). A cross-cultural study of English and Setswana speakers on a colour-triads task: A test of the Sapir-Whorf hypothesis. British Journal of Psychology, 89, 1-15.

De Valois, R.L. & De Valois, K. K. (1993). A multi-stage color model. Vision Research, 33(8), 1053-1065.

De Valois, R. L. & Jacobs, G. H. (1968). Primate color vision. Science, 162, 533-540.

Fahle, M. & Edelman, S. (1993). Long-term learning in vernier acuity: Effects of stimulus orientation, range and of feedback. Vision Research, 33, 397-412.

Fletcher, R. (1980). The City University Colour Vision Test, 2nd ed. London: Keeler.

Garro, L. C. (1986). Language, memory, and focality: A reexamination. American Anthropologist, 88, 128-136.

Gleason, H. A. (1961). An Introduction to Descriptive Linguistics. New York: Holt Rinehart & Winston.

Goldstone, R. (1994). Influences of categorisation on perceptual discrimination. Journal of Experimental Psychology: General, 123(2), 178-200.

Harnad, S. (Ed.), (1987). Categorical Perception. New York: Cambridge University Press.

Heider, E. R. (1972). Universals in color naming and memory. Journal of Experimental Psychology, 93(1), 10-20.

Hunt, R. W. G. (1987). Measuring Colour. Chichester: Ellis Horwood.

Kay, P. & Kempton, W. (1984). What is the Sapir-Whorf hypothesis? American Anthropologist, 86, 65-79.

Kay, P. & McDaniel, C. K. (1978). The linguistic significance of the meanings of basic color terms. Language, 54, 610-646.

Lucy, J. A. (1992). Language Diversity and Thought: A Reformulation of the Linguistic Relativity Hypothesis. Studies in the Social and Cultural Foundation of Language No. 12. Cambridge: Cambridge University Press.

Lucy, J. A. & Schweder, R. A. (1979). Whorf and his critics: Linguistic and nonlinguistic influences on color memory. American Anthropologist, 81, 581-615.

Newhall, S. M., Nickerson, D. & Judd, D. B. (1943). Final report of the OSA Subcommittee on the spacing of the Munsell colors. Journal of the Optical Society of America, 33(7), 385-418.

Ratner, C. (1989). A sociohistorical critique of naturalistic theories of color perception. The Journal of Mind and Behaviour, 10, 361-372.

Saunders, B. A. C. & van Brakel, J. (1997). Re-evaluating basic colour terms. Brain and Behavioural Sciences, 20, 167 228.

Sapir, E. (1921). Language. New York: Harcourt, Brace.

Whorf, B. (1956). Language, Thought and Reality. Cambridge, MA: MIT Press.
COPYRIGHT 1998 British Psychological Society
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1998 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Davies, Ian R.L.
Publication:British Journal of Psychology
Date:Aug 1, 1998
Previous Article:The role of social deviance and violations in predicting road traffic accidents in a sample of young offenders.
Next Article:Can animals detect when their owners are returning home? An experimental test of the 'psychic pet' phenomenon.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters