Universität Bielefeld - Sonderforschungsbereich 360

Isolated and Interrelated Concepts

Robert Goldstone, Indiana University

Modern research on human concept representation has evolved from two traditions. One tradition relates concepts to language (to know a concept is to be able to use a word) and has stressed how concepts are interconnected to each other. Another tradition relates concepts to object perception (to know a concept is to be able to correctly categorize perceptual inputs) and treats concepts as independent detectors. The goal of this talk is to unite these approaches by considering concepts to be both perceptually grounded and linked to other concepts. To this end, a series of experiments will be described that manipulate and measure the degree of isolation/interrelatedness of learned visual concepts. These experiments explore the categorization of prototypical and caricatured objects, and the importance of diagnostic and nondiagnostic object features for categorization. A recurrent neural network model is proposed in which concept units are connected to each other as well as to perceptual inputs, and can dynamically influence each other. This model provides an account of conceptual interrelatedness and isolation, and is applied to results obtained from psychology experiments.

Robert Goldstone is an Associate Professor in the Department of Psychology and Program in Cognitive Science at Indiana University. He is on the editorial board of Journal of Experimental Psychology: Learning, Memory, and Cognition, and has been named a Gill Fellow in Cognitive Science.

Current research interests: representation and learning of concepts (including people's acquisition of visually presented concepts, interactions between developing concepts, the influence of concepts on perception, and concept specialization/differentiation; computer simulations (often times, neural network models) to model concept formation and the two-way interaction between our conceptual and perceptual systems); judgment, decision making, and comparison processes involving similarity and analogy.


Anke Weinberger, 1997-10-24