Third European Conference on Speech Communication and Technology

Berlin, Germany
September 22-25, 1993


A Robust Analyzer for Spoken Language Understanding

Evelyne Millien, Roland Kuhn

Centre de recherche informatique de Montreal (CRIM), McGill College, Montreal, Quebec, Canada

This paper describes the CRIM Hybrid Analyzer for Natural Language, CHANEL, used as the NL component of CRIM ATIS (Air Travel Information System) system in the official DARPA November-92 evaluation. CHANEL is a robust analyzer with two components: 1) a parser based on Lexical Grammar formalism which analyzes important phrases in the sentence and finds their semantic values, and 2) a keyword classification tree (KCT) component which decides on the overall structure of the query. The system's input could be either text transcriptions of spoken sentences (NL test), or sentences output from the speech recognizer (SLS test). CHANEL converts the initial sentence into an intermediate semantic representation which is processed by another module to generate the final database query (in SQL) to obtain the information from the relational database.

Keywords: speech understanding, robust parsing, lexical grammar, keyword classification trees, ATIS.

Full Paper

Bibliographic reference.  Millien, Evelyne / Kuhn, Roland (1993): "A robust analyzer for spoken language understanding", In EUROSPEECH'93, 1331-1334.