Connectionist Symbol Processing with Causal Representations

Abstract

In this paper representation in connectionist symbol processing is addressed. There is a putative view that symbol processing requires structural representations and sensitive-to-structure procedures. Here it is proposed that symbolic information processing is based on causal nonstructural representations when computation is massively-parallel. Such representations are formed causally and need not be structural in the sense of constituent symbolic structures. A method of causal representation construction is presented along with a simple example. An implementation of the method in Simple Recurrent Network is shown.KeywordsStructural RepresentationTuring MachineSymbolic ComputationSymbol SequenceDiscrete Dynamical SystemThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.