A neural approach to concept representation suggests explanations for certain aspects of aphasia, alexia, and agraphia

Abstract

Slightly more than 10,000 constraints on permissible connection values within a relatively simple nerve network are discussed and shown to enable modeling of certain aspects of concept representation and production of semantic errors associated with injury-related aphasia, agraphia, and alexia. The constraints considered include those that must be satisfied for novel and familiar examples of familiar categories to be appropriately named, for groups containing members of several categories to be named, for members of each category to be listed, for information to be correctly transferred from one concept to another, and for the severity of errors to be minimized when inappropriate names are selected. The model is startling because it involves only six learning rules that guide adjustment of connection strengths to values which generally satisfy the 10,000 constraints, and because network connections are so strictly regulated—with several hundred constraints simultaneously limiting the range of values permitted for each connection—that it seems reasonable to ask whether any qualitatively different explanations for these aspects of concept representation are possible. This study describes a simulation rather than a mathematical model. Its inclusion in a journal of mathematical modeling is intended to spur the interest of mathematicians. so that an appropriate mathematical analysis can be conducted.