Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Multilingual Language Generation Across Multiple Domains

James R. Glass, Joseph Polifroni, Stephanie Seneff

Spoken Language Systems Group, Laboratory for Computer Science, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA

In this paper we describe a mechanism developed to generate well-formed sentences across multiple domains from a common semantic representation. This language generation component is embedded in conversational systems that permit users to query databases via spoken requests. The generation mechanism serves two distinct but overlapping roles: paraphrasing user utterances, and responding to user queries. The generation component is completely table-driven, with separate tables controlling the generation for each distinct language. Convenient mechanisms for specifying inflectional endings and for handling movement phenomena have been developed. The system can generate responses/paraphrases for a variety of languages and for several different database-query tasks. We report evaluation results for paraphrasing in an air-travel domain.

Full Paper

Bibliographic reference.  Glass, James R. / Polifroni, Joseph / Seneff, Stephanie (1994): "Multilingual language generation across multiple domains", In ICSLP-1994, 983-986.