Creating Something Different — category generation in Computational Brain & Behavior

Creating Something Different: Similarity, Contrast, and Representativeness in Categorization, with Shi Xian Liew, Nolan Conaway, and Kenneth Kurtz, appeared in Computational Brain & Behavior.

Generating a new concept is among the most striking things minds do, and most work on it has asked how the statistical structure of known categories generalizes to invented ones. This paper asks a different question: what role does contrast with the known categories play?

We propose two ways to model category contrast — one based on exemplar dissimilarity, one on the representativeness heuristic — and test them across three behavioral experiments. People do generate categories that contrast with what they have seen, distributing new examples across the unoccupied regions of the space. Which model captures them best turns out to depend on the shape of what they already know: the representativeness account wins when the known category is well described by a Gaussian, and the exemplar account takes over when it is not.