Statistical tests of cross-language color naming

Paul Kay (linguistics, University of California at Berkeley)
Terry Regier (psychology, University of Chicago)
Richard Cook (linguistics, University of California at Berkeley)
John O'Leary (computer science, University of Chicago)

It is generally accepted that there are cross-linguistic universal tendencies in the naming of colors. This is due in large part to the findings of Berlin and Kay (1969), who found universal patterns in color naming data collected from a variety of languages. Recently, however, these well-known universalist findings have been challenged, on both methodological (Lucy 1997, Saunders & van Brakel 1997) and substantive (Roberson et al., 2000) grounds. Critically, the original universalist findings are vulnerable on two key points:

  1. These findings have not yet been properly tested statistically. In Berlin and Kay (1969) a small test of three languages was performed. However, the larger claims of cross-language universals in color naming rested primarily on the intuitively apparent clustering of focal color choices for 20 languages on a discretized surface of highly saturated Munsell colors, roughly approximated by the grid shown above. Unfortunately, people sometimes perceive a seemingly non-random clustering of "hits" in items that are actually distributed randomly over a surface (Clarke, 1946). Thus, it is possible that the finding of color term universality, based as it is on subjective perception of clustering, is without statistical foundation -- a point that has not escaped critics of this work (e.g., Lucy 1997, Saunders and van Brakel 1997).
  2. The language sample from which Berlin and Kay (1969) collected data was strongly biased in favor of written languages from industrialized societies. Thus, even if their findings of color term universals had been supported by statistical test, it would still not be clear how well these results could be expected to generalize to other languages.

A study now underway at U.C. Berkeley and the University of Chicago responds to these concerns. We are statistically testing comprehensive color naming data, collected from 110 unwritten languages from non-industrialized societies, through the World Color Survey. Through these tests, we seek to establish: (i) whether color terms from different languages cluster together in perceptual color space at rates greater than chance; (ii) whether these clusters are located near the points where earlier studies of languages from industrialized societies have placed universal focal or landmark colors; and (iii) whether the color term systems of the languages studied tend to fall into a small number of distinct types forming a developmental sequence, as proposed by universally oriented work on color naming, e.g., Kay and Maffi (1999).

The color naming data on which these tests are based is publicly available on this website.

Last updated: 20030603