Semantic Category Imprecision: A Connectionist Study of the Boundaries of Word Meanings

Abstract

The view that word meanings have a discrete symbolic character has led to misleading impressions about their neurobiological basis. We hypothesize that this lexical semantic knowledge is imprecise, overlapping, and widely distributed throughout the human temporal lobes. The present study uses a computer modeling experiment to investigate this hypothesis. The objects in the study are “cups” and “bowls.” Thirty connectionist networks were taught that wide containers without handles are “bowls” and that narrow containers with handles are “cups.” The network was then asked to identify 14 different cylindrical containers of seven different widths, some with handles and others without. When compared to a study of human performance, the networks manifested the normal pattern of object naming and the normal pattern of individual variability.