|
Article Excerpt
Linguistic colour categories vary across languages (Berlin & Kay, 1969). They vary in their number of 'basic' colour categories and in their positions of category boundaries. The perception of colour is also categorical. The colour continuum is perceived as a number of qualitatively discrete categories designated in English by terms such as red, green, yellow and blue (Beare, 1963). Colour perception is also categorical in another sense. The discrimination of colours in different categories is easier (more accurate or faster) than equivalently spaced colours in the same category (e.g. Bornstein & Korda, 1984). Henceforth, this effect will be referred to as categorical perception (CP). The relationship between these two sets of colour categories (linguistic and perceptual) is a matter of debate. For some (Linguistic Relativists) the two systems are isomorphic with language constructing perceptual categories (Whorf, 1956). For others (Universalists), the two systems are less yoked: linguistic categories map onto pre-existing perceptual categories, but not all languages mark all perceptual categories.
The main concern of this paper is whether, and if so how, linguistic categories affect colour perception. (1) If the Universalist position is true, then the distribution of categorical perception across colour space should be universal. On the other hand, if the Relativist position is true, the distribution of categorical perception should co-vary with the distribution of linguistic category boundaries. Here, we report three tests of these opposing positions by comparing speakers of English and Ndonga, languages that differ in the structure of their colour categories, on colour grouping, colour triads and visual search tasks. These tasks were designed to vary in the perceptual-cognitive processes required to perform them. Thus, the profiles across tasks for each language may indicate the mechanisms producing any differences.
Physiology of colour categories
A great deal is known about the physiological basis of colour vision, from genes through the biochemistry of the retinal receptors, to the early stages of neuronal processing (e.g. Mollon, Pokorny, & Knoblauch, 2003). However, the physiological basis for colour categories remains unknown. Early work (De Valois & Jacobs, 1968) seemed to have discovered the basis for the perceptual primacy of red, green, yellow and blue (Hering's unique hues). Kay and McDaniel (1978) suggested that the unique hues functioned as category prototypes around which the four primary categories formed. Secondary categories, such as orange, were formed by combination of two primary categories (in this case red-yellow). However, De Valois and Jacobs (1968) interpretation of their single cell recordings has had to be revised (De Valois & De Valois, 1993). These cells seem to code, for instance, magenta and cyan, rather than red and green, and the physiological basis for unique red and green is still not known (Jameson & D'Andrade, 1997). There have been other notable attempts to provide a physiological account for colour categories (e.g. Ratliff, 1976), but they lack independent corroborative support (Abramov, 1997).
Innateness of colour categories?
At a behavioural level, there is strong evidence that colour categories are present before language and they may be innate (Bornstein, Kessen, & Weisskopf, 1976; Catherwood, Crassini & Freiberg, 1989). Infants at four months, following habituation to one colour (say blue 1) dishabituated to a novel colour provided it was from a new adult category (green) but not to a novel same-category colour (blue2). This same pattern was shown for all boundaries among the primary categories (blue-green, green-yellow and yellow-red), and Franklin and Davies (2003) found that this categorical responding was also found for secondary boundaries such as red-pink and blue-purple. These studies provide strong evidence consistent with some degree of innateness of colour categories. However, the infants in both studies behaved very differently to adults when they show categorical perception. The infants showed no interest (i.e. no dishabituation or preferential looking) if the novel colour was from the same category, even though the difference between the original colour and the novel colour was above threshold. The equivalent to adult categorical perception would have been for the infant to show some dishabituation or preference for the novel within-category colour, but more dishabituation or preference for the new-category colour. It is possible that other methods would induce the latter pattern, but as it stands, it is not clear whether infant colour categories are directly equivalent to adult colour categories, or not.
Even if colour categories are innate and universal, this does not preclude the possibility that they are modified by later experience, as seems to be the case with in speech perception (Werker & Tees, 1984). For instance, categorical effects could be strengthened if they are reinforced by language marking the distinction, and they may also weaken if they are not supported by linguistic categories.
Tests of linguistic relativity
The first tests of the linguistic relativity hypothesis (LRH) used the method of recognition memory and found that the more nameable or codeable the colour the easier it was to remember (Brown & Lenneberg, 1954; Lenneberg & Roberts, 1956; Stefflre, Vales, & Morely, 1966). These findings were taken as supporting the LRH through the language providing effective codes for remembering particular stimuli. In our terms, this is a 'direct-language' explanation. The benefit arises from labelling the target colour and retaining the name as an aid to recognition. However, the easily remembered colours might also have been perceptually distinctive, and this made them both easier to remember and to label. Such distinctiveness could be either hardwired and universal, or induced by language learning. However, as these early studies did not include cross-language comparisons, the possibility that some colours were distinctive independently of language could not be tested.
Rosch-Heider and colleagues explored the above possibilities in a classic series of experiments (Rosch-Heider, 1972; Heider & Olivier, 1972). They compared the colour memory of English-speaking Americans and the Dani, whose language had just two basic colour terms. Despite the Dani's simple colour language, the Dani, like the American-English, found focal examples of Berlin and Kay's universal categories easier to remember than poorer examples (Rosch-Heider, 1972). Moreover, the pattern of confusion errors was also found to be similar between groups (Heider & Olivier, 1972). The results were attributed to the universal perceptual salience of the focal colours and were subsequently taken as settling the issue in favour of Universalism.
Rosch-Heider's methods and her interpretation of her findings were not without criticism (e.g. Lucy & Schweder, 1979, Ratner, 1989; Ratner & McCarthy, 1990, Simpson, 1991). However, most problematic for the Rosch-Heider studies is a recent series of experiments by Roberson, Davies and Davidoff, (2000). Rosch-Heider's methods were replicated in a comparison of English speakers and speakers of a five-basic-term language, Berinmo (like Dani a language of the Melanesian family). However, unlike Heider & Olivier (1972), patterns of naming correlated positively with memory confusions within each language (Roberson et al., 2000, Expt 1a). Roberson et al. pointed out that Rosch-Heider also found the latter pattern, but she chose to ignore this evidence in favour of the evidence from multidimensional scaling of the memory confusions that showed that the English and Dani solutions were very similar. Note, though, that this judgment was based on subjective visual comparison with no formal statistical support. The Berinmo also showed better recognition memory for focal colours over non-focal colours (Roberson et al., 2000, Expt 2a), as Rosch-Heider had found, but once response bias in responding had been allowed for, this advantage disappeared. Thus, although Roberson et al.'s results were broadly similar to Rosch-Heider's, different statistical techniques suggested that the data were consistent with a degree of Relativism.
The case for Relativism was strengthened by a further experiment that compared CP in the two groups (Expt 6a). Berinmo has a combined term for blue and green and thus has no blue-green boundary. On the other hand, the position of the Berinmo nol-wor boundary has no equivalent in English. Discrimination accuracy across these two boundaries was compared to within-category discrimination using a successive 2-AFC method. Following presentation of a target colour, two test colours were presented after a five-second delay. One test colour was the target and the distractor was either in the same category as the target (e.g. blue 1 target, blue 2 distractor) or was from the adjacent category (e.g. blue 1, green 1) for the language-boundary tested. English speakers showed categorical perception for the blue-green boundary, but not for the nol-wor boundary, whereas the Berinmo showed the opposite pattern.
Direct or indirect language effects?
Most tests of the LRH, including Roberson et al.'s, involve a memory component and are particularly prone to direct language strategies (Kay & Kempton, 1984; Pilling, Wigget, Ozgen, & Davies, 2003; Roberson & Davidoff, 2000). For instance, the target colour could be remembered as a colour name and recognition achieved by matching the retained name to the names of the colours in the recognition array. For many combinations of target and recognition array, a naming strategy confers an advantage on those with the largest set of available names. The latter could therefore explain why the Dani (Rosch-Heider, 1972) and the Berinmo (Roberson et al., 2000) were considerably worse at the recognition memory task than the English groups. Moreover, this would produce the observed correlations between naming and errors. Although less explicitly a memory task, this direct language account can be extended to Roberson et al.'s finding of different patterns of categorical perception for Berinmo and English. Cross-category discrimination could (in part at least) be supported by the stimuli having different names, whereas within-category stimuli cannot. This naming advantage would be available to the English for the blue-green boundary, but not for the nol-wor boundary, and the reverse would be the case for the Berinmo.
This direct language account of CP was supported by a within-language study (Roberson & Davidoff, 2000). The same method as outlined earlier was used, and CP for the blue-green boundary was tested. In the baseline condition (unfilled ISI) cross-category discrimination was more accurate than the equivalent within-category discrimination. However, this cross-category advantage was eliminated...
|
