Feature encoding and pattern classifications with sequentially presented Markov stimuli

The major objective of this experiment was to develop and evaluate a methodology designed to permit more direct assessment of the detailed processes involved in prototype abstraction. Thirty Ss participated in a task having the following characteristics: (1) classifications of Markov-generated stimuli sampled from two different populations, (2) controlled scanning of pattern features, (3) a measure of the degree to which pattern features were correctly identified, and (4) intermittent reproduction of pattern features abstracted from collections of mixed instances. Results showed that a significant number of the Ss learned to classify the stimuli into categories corresponding to generation rules and, at least partially, abstracted the population prototypes from these variable instances. The feature identification data suggested that the Ss who were unsuccessful in classifying the stimuli into the rule-defined categories used an inappropriate strategy for sampling pattern information upon which to base their classifications.