Categorization is a fundamental cognitive ability both for humans and other animals. Characterizing and quantifying the degree of difficulty of category learning is important to understanding how the brain categorizes. It is known that different categories are learned with varying degrees of difficulty. The present paper seeks to elucidate, using a neural net model of experimental outcomes in categorization tasks, (1) what makes some categories more difficult to learn than others and (2) how does task difficulty relate to categorical perception (CP), the phenomenon in which the internal representation of similarities is modified by category learning such that inputs belonging to different categories come to be perceived as more different after the category is learned and inputs belonging to the same category come to be perceived as more similar. Four parameters are defined in this paper to control and evaluate category learning difficulty. We show how these parameters relate to category-learning difficulty and in turn provide clues as to how category-learning difficulty may be related to CP.
Keywords: categorization, complexity, categorical perception, neural net.