Constraints on Colour Category Formation

Abstract

This article addresses two questions related to colour categorization, to wit, the question what a colour category is, and the question how we identify colour categories. We reject both the relativist and universalist answers to these questions. Instead, we suggest that colour categories can be identified with the help of the criterion of psychological saliency, which can be operationalized by means of consistency and consensus measures. We further argue that colour categories can be defined as well-structured entities that optimally partition colour space. We provide some empirical support for this claim by presenting experimental results, which indicate that internal structure is a better predictor of colour categories than perceptual saliency.

Acknowledgements

This study was conducted with funding from the University of Groningen. We would also like to thank the Centre for Computation, Cognition, and Culture, Psychology Department, Goldsmiths, University of London, where the studies described in this article were carried out; the Fyssen Foundation for the postdoctoral grant allocated to the first author; the Centre for the Study of the Senses, Institute of Philosophy, School of Advanced Study, University of London, for having hosted the first author throughout the academic year 2010/11; Jules Davidoff for his help with the experimental set-up, feedback, and comments on the earlier version of this paper; Barry C. Smith for his advice on this study; Nicolas Claidière for his participation to the design of this study. An earlier version of this paper was presented at the 2011 Philosophy of Science Conference at the Inter-University Centre, Dubrovnik.

Notes

In speaking of equidistant colours, we are assuming that colours can be represented geometrically. In fact, we are taking for granted that colours can be represented in a metrical colour space, such as the CIELab or CIELuv space. See below for more on the notion of colour space.

To avoid confusion in this paper, and to distinguish between cases where we refer to colours from cases where we refer to colour terms and categories, we place the latter within single quotes. We trust that context will disambiguate between colour terms and colour categories.

Some authors (e.g. Berlin and Kay 1969; Rosch 1973; Boynton and Olson 1987) construe psychological saliency as also involving quick response times. Our operational definition deviates from this use because, motivated by the studies discussed in the previous section, we want to distinguish between perceptual and psychological saliency.

Given that we define psychological salient categories on the basis of consensus and consistency, it follows that colour categories, the members of which have a high degree of membership, are psychologically salient.

It is worth noting that in studies using RTs to establish perceptual saliency in pop-out effects and visual search tasks, target colours are presented among distracter colours. In our study, the target colour is presented alone on the screen while the participant judges whether or not the sample is a member of a given category.

We expected the average RT in the ‘no’ end points of the ‘blue’, ‘green’, and ‘purple’ conditions to be significantly faster than the average RT across responses, as in the cases of ‘brown’ and ‘lilac’. A possible explanation of why this was not observed is linked to the fact that ‘blue’, ‘green’, and ‘purple’ are the categories that have the widest extensions in English partitioning of colour space, along with ‘pink’ perhaps. It is possible that the samples chosen beyond these categories' borders did not go far enough for us to be able to observe a consistent decrease of RTs in the ‘no’ response. Although this needs further testing to be confirmed, we can fairly confidently state that had samples been chosen further out of the centre, we would have observed the same kind of behaviour for the reaction times at the end points of ‘blue’, ‘green’, and ‘purple’, as in the case of ‘brown’ and ‘lilac’.

Colours presented in these tasks varied in hue and lightness. In the presentation of the results, the lightness dimension was collapsed into the hue dimension, such that data points on each side of the category centre represent colour samples that vary in both dimensions, not just in hue. Thus, colours from all around a given category centre were in fact presented in these tasks.