A Relaxation Method for Understanding Speech Utterances

A Relaxation Method for Understanding Speech Utterances 1 Stephanie Seneff Spoken Language Systems Group Laboratory for Computer Science Massachusetts Institute of Technology Cambridge, Massachusetts 02139 U.S.A. Spontaneous ABSTRACT This paper describes an extension to the MIT ATIS (Air Travel Information Service) system, which allows it to answer a question when a full linguistic analysis fails. This "robust" parsing capability was achieved through minor extensions of pre-existing components already in place for the full linguis- tic analysis component. Robust parsing is apphed only after a full analysis has failed, and it involves the two stages of 1) parsing a set of phrases and clauses, and 2) gluing them together to obtain a single semantic frame encoding the full meaning of the sentence. We have assessed the degree of suc- cess of the robust parsing mechanism through a breakdown of the performance of robustly parsed vs. fully parsed sentences on the October '91 "dry-run" test set. It was clear that the robust parser allowed us to answer many more questions cor- rectly, as over a third of the sentences were not covered by the grammar. We also report here on the performance of the system on the February '92 test sentences, and discuss some issues with regard to the evaluation methodology. INTRODUCTION Current approaches to the language understanding aspect of spoken language systems tend to fall into two categories. In syntax-driven formulations [1,4,10], a com- plete syntactic analysis is performed which attempts to account for all words in an utterance. While provid- ing strong linguistic constraints to the speech recogni- tion component and a useful structure for further lin- guistic analysis, such an approach can break down in the presence of unknown words, novel linguistic constructs, recognition errors, and some spontaneous speech events such as false starts. In contrast, semantic-driven ap- proaches [2,5,9] tend to derive their understanding by spotting key words and phrases in the utterance. While this approach can potentially provide better coverage and deal with ill-formed sentences, it provides less constraint for the speech recognizer, and may not be able to ade- quately interpret complex linguistic constructs. This paper describes our efforts to develop a language understanding component that combines the advantages of both of these approaches. Our strategy has been to 1This research was supported by DARPA under Contract N00014-89-J-1332, monitored through the Office of Naval Research. 299 relax the constraint that the syntactic analysis must ac- count for all of the words in an utterance. Our current implementation is a two stage process. In the first step, our parser [7] searches for a complete linguistic analysis. Failing that, constraints of the parser are relaxed to per- mit the recovery of parsable phrases and clauses within the sentence. These fragments are fused together using a mechanism that closely resembles our discourse history mechanism [8]. Thus the robust parser is able to leverage off of existing components to a large degree. ROBUST PARSING MECHANISM The natural language component of the MIT ATIS system makes use of a semantic frame …

Content truncated to 500 words