Taboada, Maite
Carnegie Mellon University and Universidad Complutense de Madrid

Maite Taboada
238 Cyert Hall
Center for Machine Translation
Carnegie Mellon University
Pittsburgh, PA 15213

Ph (412) 268-6594
Fax (412) 268-6298

Strand: computational / discourse English

Title: Contextual Information for Disambiguation in a Speech-to-Speech Translation System

For any given utterance out of what we can loosely call 'context', there is usually more than one possible interpretation. A speaker's utterance of an elliptical expression, like the figure "twelve fifteen", might have a different meaning depending on the context of situation, the way the conversation has evolved until that point, and the previous speaker's utterance.

In this paper I will explore those three strata of discourse in order to gain on understanding of how speakers 'mean' in a given situation. This work is part of the JANUS multi-lingual speech-to-speech translation system designed to translate spontaneous dialogue in a limited domain (Lavie et al. 96). JANUS is designed to deal with the kind of problems that naturally occur in spontaneous speech --- such as mispronunciations, restarts, noises, slightly ungrammatical input, and the lack of clear sentence boundaries --- with additional errors introduced by the speech recognizer. The machine translation component of JANUS handles these problems using two different approaches: GLR* and Phoenix. The GLR* parser (Lavie and Tomita 93) is designed to be more accurate, whereas the Phoenix parser (Ward 91) is more robust. Both are language-independent and follow an interlingua-based approach. The current system translates spontaneous dialogues in the scheduling domain, with English, Spanish, and German as both source and target languages.

This project addresses the problem of choosing the most appropriate semantic parse for any given input. The approach is to combine discourse information with the output of the Phoenix parser, a set of possible parses for an input string. There might be more than one acceptable semantic parse for an input. The discourse module interacts with the parser, selecting one of these possibilities. The decision is to be based on:

The context module in the system keeps a global history of the conversation, from which it will be able to estimate, for instance, the likelihood of a greeting once the opening phase of the conversation is over. A more local history predicts the expected response in any adjacency pair, such as a question-answer sequence.


[Halliday 94] M.A.K. Halliday. An Introduction to Functional Grammar, Edward Arnold, London, 1994 (2nd edition).

[Lavie et al. 96] A. Lavie, D. Gates, M. Gavalda, L. Mayfield, A. Waibel, L. Levin. Multi-lingual Translation of Spontaneously Spoken Language in a Limited Domain. In Proceedings of COLING 96. Copenhagen. 1996.

[Lavie and Tomita 93] A. Lavie and M. Tomita. GLR*: An Efficient Noise Skipping Parsing Algorithm for Context Free Grammars. Proceedings of the Third International Workshop on Parsing Technologies, IWPT 93, Tilburg, The Netherlands, 1993.

[Martin 92] J. Martin. English Text: System and Structure. John Benjamins. Philadelphia/Amsterdam. 1992.

[Schegloff and Sacks 73] E. Schegloff and H. Sacks. Opening up Closings. Semiotica 7, 289-327, 1973.

[Ward 91] W. Ward. Understanding Spontaneous Speech: the Phoenix System. In Proceedings of ICASSP 91, 1991.