Symbols Among the Neurons: Details of a Connectionist Inference Architecture

Abstract

Pattern matching a:1d variable binding are easily implemented in conventional computer architectures, but not necessarily in all architectures. In a distributed neural network architecture each symbol is represen ted by activity in many units and each unit contributes to the representation of many symbols. Manipuiatirg symbols using this type of distributed repres c nta!i~1n is not as easy as with a local representation wh8re each unit denotes one symbol , but there is evidence that the distributed approach is the one chosen by nature. We describe a v;orking implementation of a production system interpreter in a neural network using distributed representations for both symbols and rules. The research provides a detailed account of two important symbolic reasoning operations, pattern matching and variable binding, as emergent properties of collections of neu ron -like elements. The success of our produc tion system implementation goes some way towards c>.nswering a common criticism of connectionist theories: that they aren't powerful enough to do symbolic reasoning.