Toward a framework for combining connectionist and symbol-processing models

Abstract

The new connectionist models of cognition represent a convergence of trends originating in the study of sensory systems, control theory, and neoassociationism. The strengths of these models correspond in a complementary way to weak points in current symbol-processing theory, suggesting possibilities of fruitful combination or collaboration. Some specific routes to these ends are considered with specific illustrations from the domain of category learning. In particular, it appears that a connectionist network model for learning might fruitfully be coupled to a memory array model for the judgmental aspects of categorization. Considered more broadly, connectionist models raise issues of a generality/specificity trade-off, issues of sufficiency and testability, and questions about the relationship between models and levels of information processing.