Abstract
The basic colour terms of a language are the smallest set with which a speaker can name any colour. Based on a study of 20 languages it has been claimed that the “basic colour terms of any given language are always drawn” from a set of eleven: black, grey, white, red, orange, yellow, green, blue, purple, pink & brown. Although this claim to cultural universality has been contested through criticism of methodology and identification of possible counter examples, it has recently been confirmed by analysis of data from the World Colour Survey. I have tested the hypothesis that the basic colour categories (BCCs) are uniquely optimally effective for describing the colours of everyday things. Effectiveness was measured by success at a classification task: given colour based descriptions of three things, two of which are from the same class (e.g. two people) while the third is from another class (e.g. a tree), spot the odd one out. I used classes defined by 758 concrete nouns. For instances of each class I used the first 80 images return by ‘Google Image’ in response to a query using the class-defining noun. Candidate systems of categories were partitions of the RGB cube compatible with a 4*4*4 quantization. Colour-based descriptions (relative to an RGB partition) were histograms of the proportion of image pixels within each colour category. Guessing the odd one-out was done by eliminating the most similar pair of descriptions, where similarity was measured by Euclidean distance in a whitened space of square-rooted histograms. A subset of BCC-like partitions was defined from previously naming data. The results showed that the best BCC-like partition was better than chance and no other partition was significantly better than it. Out of a class of qualitatively similar systems of categories obtained by rotating the BCCs about a lightness axis, the BCC-like partition was the best. These findings support the hypothesis that the BCCs are universal because they are optimal.