Abstract
Categorical perception, or more precisely the associated phenomenon known as "acquired distinctiveness" (AD), refers to the enhancement of perceptual sensitivity along category-relevant or "informative" perceptual features as a result of category training. But exactly which perceptual features the system treats as "informative" is not well understood, in part because almost all studies to date have used categories separated by a hard boundary, in which only one feature---the one crossing the boundary---is informative by any reasonable definition. This study aims to investigate more comprehensively which perceptual features enjoy an improvement in perceptual sensitivity as result of category training. In a series of experiments, subjects learned to distinguish two shape categories defined as Gaussian distributions in an unfamiliar 2-dimensional feature space. These categories define a "soft" optimal linear boundary, whose precision could be controlled by modulating the category overlap. Shape features that are orthogonal to this boundary are maximally informative about category membership; features that are parallel to the boundary are completely uninformative; and other features (e.g. "diagonal" features) have intermediate levels of informativeness. The degree of informativeness of each feature can be defined as the mutual information between the feature and the category variable, which is the degree to which knowledge of the feature reduces Shannon uncertainty about the category. Perceptual discrimination was tested before and after category learning at various features in the space, allowing the magnitude of AD attributable to training to be measured. The results support a remarkably simple generalization: the magnitude of improvement in perceptual discrimination (AD) at each feature was proportional to the mutual information between the feature and the category variable. This finding suggests a "rational" basis for categorical perception, in which the precision of perceptual discrimination is tuned to the statistical structure of the environment.