Application of the Bower and Trabasso theory to four-category concept learning with probabilistic feedback

Abstract

Four-category conjunctive concept problems were learned under nine different conditions with the number of irrelevant dimensions and the percentage of correct feedback as independent variables. Concept identification performance could be adequately predicted by assuming that the two relevant dimensions defining the conjunctive problems were learned concurrently and independently. A generalized version of the Bower-Trabasso theory of concept identification, encompassing a distinct response learning phase, was applied to the data. The theory adequately described performance under conditions with consistently correct feedback. Several alternative models for learning under misinformative feedback conditions were developed from the theory. Using parameter estimates from data under correct feedback conditions, these models were found to predict performance under misinformative feedback fairly well.