Association‐based parser for speech understanding systems—details of knowledge and performance evaluation

Abstract

AbstractIn language processing aiming at speech understanding, not only knowledge about the syntax and the semantics, but also knowledge about the context and the control play important roles in dissolving the ambiguity due to the uncertainty in the phoneme recognition. This paper discusses the association‐based parser ASP. Discussions are made especially on the details of the employed knowledges and the result of evaluation experiment, which was performed to verify the usefulness of the association mechanism. In ASP, based on the semantic memory, episode memory, syntactic knowledge and control knowledge, the vocabulary associated with a topic in the sentence is processed with priority, and the control of the search in the intelligent backtrack is performed. In the evaluation experiment, the control knowledge to determine the object of processing from the contending partial solutions, as well as the effect of the order of solution candidates in the phrase, are particularly examined. It was verified as the result that ASP exhibits a performance independent of the order of the solution candidates under a pertinent control strategy, and the correct phrase sequence is identified, based largely on the syntactic and semantic informations, even if there does not exist an appropriate topic.