RULE PERFORMANCE AND GENERALIZATION IN A MATCHING-TO-SAMPLE TASK

This study examines the relationship between human subjects' self-generated rules and their performances in a matching-to-sample task. Twenty subjects were randomly divided into two training groups which were given unequal amounts of stimuli dimensions to take into account. All subjects were trained in a matching task using geometrical figures until they reached 100% performance. At the end of each session they were asked to describe the key to performance and try to form a more general rule. Once the performance criterion had been achieved, all the subjects went on to a generalization session, in which we found that only those who had made up a general rule were able to perform the task correctly. The type of training did not have a significative effect on both the type of self-reports produced by the subjects and their performances.