Efficient Semantic Constraint for Speech Recognition

Abstract

In this paper we present a method for describing semantic constraint in SUSKIT-II, the speech understanding system we are developing. We integrate both syntactic and semantic constraints in a definite clause grammar. The syntax is expressed by a set of rewriting rules and the semantics by arguments of nonterminal symbols and auxiliary predicates on some arguments. These arguments, called semantic parameter, detail words semantically. An auxiliary predicate is a relation among semantic parameters to describe the simultaneous occurrence of semantic word classes. Moreover, using the relation, we introduce a concept hierarchy on the domain formed by a set of values which each semantic parameter can take. Since unnecessary resolution in the semantic level is avoided by means of this device, an amount of memory to store the linguistic structures of intermediate hypotheses could be reduced by 78.5 % in comparison with the case that the concept hierarchy was not used.