An alternative neural network representation for conceptual knowledge

Abstract

This paper introduces the P-Graph representation of a neural network as an alternative to the classical « semantic networks » introduced in knowledge representation by Quillian. None of the shortcomings of Quillian-type semantic networks are displayed by it. The P-Graph is a particular type of dual of a graph: memory traces (typically “words”) are associated with the edges of the graph, the relations between the memory traces, with the vertices. The P-Graph is the mathematical object underlying ANELLA (Associative Network with Emergent Logical and Learning Abilities). The P-Graph – in particular the way it grows - is shown to be compatible with the architecture of an actual biological neural network, its emergent logical and learning abilities are shown on examples borrowed from the working of ANELLA as developed at British Telecom Laboratories in 1988 under a BT Academic Fellowship.