The perceptual space that results from the processing of opponent colours is non-uniform (see Figure below), meaning that there are optimal ways to describe it (Jameson & D’Andrade, 1997). See this excellent tutorial for more details.
Natural language partitions are optimised for describing the perceptual colour space, suggesting that perception is a primary constraint on colour categorisation (Regier, Kay & Khetarpal, 2006). This was concluded from a model which partitioned the Munsell chip colour set so as to optimise the ‘well formedness’ of the partitions. Well-formedness is a measure of how the partition of colour space maximises similarity within a category and to what extent words tend to name connected regions. The first constraint biases categories to be regularly shaped rather than cover large ranges of the colour space. The second constraint biases categories to cover a single, connected space, rather than isolated patches. The model’s output resembled natural human partitonings (from the WCS). Furthermore, for 82 of the 110 natural languages in the WCS, the well-formedness decreased as their partitions were shifted.
This mirrors findings in the configurations of vowel spaces (e.g., Liljencrants & Lindbloom, 1972, De Boer, 2000). Taken on its own, this evidence would support a Universalism, although later sections will show more complex interactions.
Buchsbaum and Bloch (2002) found that similar algorithms approximate low-level processing of colour in the striate cortex (red-green and blue-yellow opponent contrasts, by Principal Component Analysis, PCA) and high-level linguistic encodings of colour (colour terms, by Non-negative Matrix Factorisation, NMF). NMF is a factor analysis algorithm like PCA, except that it is designed for values that are inherently positive and cannot be centred. NMF has been applied to pictures of faces and has been shown to isolate features such as noses, eyes, ears (Lee & Seung, 1999).
Colour terms also emerge from the latter in a similar order to Berlin & Kay’s universal colour order. NMFs can be manipulated to extract any number axes to describe the data (basis functions). In this context, the number of basis functions requested is analogous to the number of colour terms in the language. NMFs were derived from the Munsell colour space with 3, 4, 6 and 8 basis functions, then each basis function was translated to an English colour term using data from the WCS. The addition of colour terms as the number of basis functions increased, although not deterministic, generally followed Berlin and Kay’s (1969) universal order. That is, labels for colours emerge first for areas of the space which are easier to distinguish cognitively.
Although this may be evidence for innate perceptual biases, it also shows that perceptual and conceptual processing are similar, which is predicted by Embodied Cognition. Buchsbaum and Block also point out that participants may be optimally dividing the Munsell chip space rather than describing their natural colour categories, confounding some of the findings for perceptual constraints.
Jameson, K., & D’Andrade, R.G. (1997). It’s not really Red, Green, Yellow, Blue: An Inquiry into cognitive color space Color Categories in Thought and Language DOI: 10.1017/CBO9780511519819.014
Regier, T., Kay, P., & Khetarpal, N. (2007). Color naming reflects optimal partitions of color space Proceedings of the National Academy of Sciences, 104 (4), 1436-1441 DOI: 10.1073/pnas.0610341104
Liljencrants, J., Lindblom, B., & Lindblom, B. (1972). Numerical Simulation of Vowel Quality Systems: The Role of Perceptual Contrast Language, 48 (4) DOI: 10.2307/411991
DEBOER, B. (2000). Self-organization in vowel systems Journal of Phonetics, 28 (4), 441-465 DOI: 10.1006/jpho.2000.0125
Buchsbaum, G. (2002). Color categories revealed by non-negative matrix factorization of Munsell color spectra Vision Research, 42 (5), 559-563 DOI: 10.1016/S0042-6989(01)00303-0