ABSTRACT
In this chapter my goal is to provide a brief introduction to two topics in knowledge representation that have attracted particular interest in recent years. The first goes under the various names of “connectionism”, “parallel distributed processing”, and “neural networks”, and refers to the idea that certain mental representations are distributed. The meaning and significance of this term will become clearer below, but the essential idea is that much of the traditional language of cognition (“symbols”, “rules”, “modularity”, etc.) should be abandoned in favour of a style of theorizing which concentrates on complex interactions between patterns of excitatory and inhibitory activation distributed across many simple processing elements. As anybody interested in knowledge representation will appreciate, recent work on connectionist networks has raised a number of controversial issues concerning mental representation. In this chapter I shall briefly review some of these issues: are distributed representations fundamentally different from the traditional symbolic representations assumed in language-of-thought theories of cognition (and if so, how)? Are distributed representations adequate from a cognitive point of view? And what do connectionist systems have to say about explicit rules?
