Connectionism Versus Symbolism in High-Level Cognition

Symbolic processing rests on a computational technology that includes dynamic memory management, virtual pointers to created structured objects, and the use of variables to propagate bindings. Connectionist models have lacked these features, but create distributed representations that are “rich,” i.e., that encode numerous statistically based expectations, acquired from experience. It is argued here that a synthesis of both symbolic and connectionist features will make important contributions to our understanding of high-level cognition. In what follows, a motivation is given for the need of such a synthesis, along with several, novel techniques used in connectionist systems designed to perform high-level, language-related processing tasks.