Parsing of Spoken Language under Time Constraints

Abstract

Spoken language applications in natural dialogue settings place serious requirements on the choice of processing architecture. Especially under adverse phonetic and acoustic conditions parsing procedures have to be developed which do not only analyse the incoming speech in a time-synchroneous and incremental manner, but which are able to schedule their resources according to the varying conditions of the recognition process. Depending on the actual degree of local ambiguity the parser has to select among the available constraints in order to narrow down the search space with as little effort as possible. A parsing approach based on constraint satisfaction techniques is discussed. It provides important characteristics of the desired real-time behaviour and attempts to mimic some of the attention focussing capabilities of the human speech comprehension mechanism.