Abstract
In sequential color sorts, subjects partition a palette of diverse color samples into n-piles (n=2…N). The composition of the piles often approximates the basic color categories associated with Berlin & Kay’s theory of color term evolution. Here, we show that this result depends on the to-be-sorted colors. Subjects saw a randomly-ordered array of iPad color samples and “slid” them into virtual “bins”. We obtained the standard result when 44 subjects sorted a 25-color palette (“P1”) that varied smoothly in hue at approximately constant Munsell value (lightness), and contained good examples of the basic color categories. However, we obtained something quite different when 73 subjects sorted a similarly-sized palette (“P2”) that sparsely sampled Munsell color space, with multiple values for many hues, while still containing good examples of the basic categories. P2 subjects most frequently sorted by value or by complex combinations of hue and value. We then obtained sequential pile-sorts (n=2…6) of palette “P3”, which varied smoothly in hue, with one random Munsell value for each hue. For P3, the iPad tracked the order in which subjects populated the bins. In aggregate, 49 subjects’ P3 results were indistinguishable from the P1 results. We propose that pile-sorts are mediated by processes analogous to Ensemble Coding, whereby subjects plan pile-sort strategies based on pre-determined color categories, prior to executing an n-sort. Consistent with this hypothesis, P3 subjects often populated the bins sequentially (43% of piles), placing most samples belonging to one category in one bin, before moving on to place the next category of samples into another bin. This result is inconsistent with the proposal that subjects generally adopt an ad-hoc strategy based on the pairwise similarities among color samples. Unlike traditional categorical parsing of colors by hue, our results also highlight the importance of the covariation of both hue and lightness.