Methods of Large Grammar Representation in Massively Parallel Parsing Systems

This paper describes techniques for massively parallel parsing where sequences of lexical categories are assigned to single processors and compared in parallel to a given input string. Because even small grammars result in full expansions that are much larger than the largest existing massively parallel computers, we need to develop techniques for "doubling up" sequences on processors so that they don't interfere during parallel matching. This paper describes three such techniques: (1) discrimination by length, (2) discrimination by open class/closed class words, and (3) combined discrimination by length and word class. We discuss possible reductions of the sequence space and implementation techniques on a CM-5 Connection Machine".