| This work focused on testing the role of selective attention in rule-based category learning tasks as instantiated in the explicit system of the COVIS (Competition between separate Verbal and Implicit Systems) model of category learning (Ashby, Alfonso-Reese, Turken, & Waldron, 1998). The first two experiments manipulated the nature of shifts in selective attention (e.g., switching attention away from a rule or selecting a new rule to attend to) required when learning two consecutive categorization tasks that could differ in the relevant and/or irrelevant dimensions. Experiment 3 introduced a novel paradigm designed to allow the decision strategy to be observable on a trial-by-trial basis in order to track how reinforcement history influences shifts in selective attention. The results of the empirical work suggested that: (1) both weakening of the irrelevant dimension and strengthening of the relevant dimension are important for category learning; and (2) people may not naturally be predisposed to persevere in categorization tasks. These data were used to test a novel computational implementation of the COVIS explicit system. In addition to supporting the empirical results, the computational work: (1) reinforced the idea that differences in extradimensional and intradimensional shifts of attention should be reflected in models of category learning; (2) suggested that COVIS should be modified to maintain a representation of the rule history, as well as the current set of vi active rules, in working memory; and (3) suggested that the rule selection and salience updating mechanisms in COVIS are critical, but that the rule switching mechanism may not be necessary under typical task conditions. |
