Learning and prototype formation of artificial categories defined by multidimensional normal distributions

In order to approximate the variety of natural categories, two kinds of stimuli whose attributes vary according to different multivariate normal distributions are generated on CRT screen. Twenty undergraduate and graduate students served as subjects. Subject's task was to discriminate the stimuli. Results revealed that subject's performance can be explained very well by the parameters of the distributions (Exp.1). Although in previous studies prototypes have been assumed to be the mean value or central tendency of category instances, there is another possibility, especially when two or more concepts are simultaneously learned, that they have attributes which are emphasized. This hypothesis was confirmed in Exp. 2, using 30 undergraduates as subjects. It was argued that we must distinguish at least two kinds of prototypes : the one formed by the most frequent instances and the other the most discriminate in contrast with the other concepts.