Evaluating categorization and connectionistic models of conceptual rule learning
Abstract
Three current categorization models, ALCOVE (Kruschke, 1992), the configural-cue model (Gluck & Bower, 1988b), and RULEX (Nosofsky, Palmeri, & McKinley, 1994) were rigorously tested on a rich data set collected in an experiment, which employed the conceptual rule learning paradigm. Subjects were given relevant attributes and dimensions and were asked to learn an unspecified rule that relates the attributes. The main results of this study were consistent with previous findings, the conjunctive the easiest and the biconditional the hardest; but allowed much finer tests of the models. ALCOVE was superior to the other models not only in quantitative fitting of learning curves of the eight fundamental rules but in qualitative predictions regarding learning curves of logical subgroups of stimuli. In qualitative predictions of transfer between rules, ALCOVE was as good as the configural-cue model, although ALCOVE provided the best quantitative fit for transfer performance. Possible reasons of the successes and failures of the models were examined.
Degree
Ph.D.
Advisors
Busemeyer, Purdue University.
Subject Area
Psychology|Experiments|Cognitive therapy
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.