Representational and architectural issues in a limited-domain medical speech translator

TALN 2005, Dourdan, 6–10 juin 2005 Representational and architectural issues in a limited-domain medical speech translator Manny Rayner (1), Pierrette Bouillon (1), Marianne Santaholma (1), Yukie Nakao (2) (1) University of Geneva, TIM/ISSCO 40, bvd du Pont-d’Arve, CH-1211 Geneva 4, Switzerland mrayner@riacs.edu, Pierrette.Bouillon@issco.unige.ch, Marianne.Santaholma@eti.unige.ch (2) National Institute for Communications Technology, 3-5 Hikaridai, Seika-cho, Soraku-gun, Kyoto, Japan 619-0289 yukie-n@khn.nict.go.jp Mots-clefs : reconnaissance de la parole, traduction de la parole, aide au diagnostic médical Keywords: speech understanding, speech translation, computer-aided diagnosis Résumé Cet article dresse un aperçu du système MedSLT, un système de traduction de la parole dans le domaine médical pour un vocabulaire limité. Il met l’accent sur le problème du choix du type de représentation pour les constructions temporelles et causales. Nous montrons que celles-ci ne peuvent pas être représentées par des structures plates, généralement utilisées pour ce type d’application, mais qu’elles nécessitent des stuctures plus riches, enchâssées, qui permettent d’obtenir une traduction plus adéquate. Nous expliquons comment produire ces représentations et écrire des règles de traduction économiques qui mettent en correspondance les représentations sources dans la représentation interlingue correspondante Abstract We present an overview of MedSLT, a medium-vocabulary medical speech translation system, focussing on the representational issues that arise when translating tem- poral and causal concepts. Although flat key/value structures are strongly preferred as semantic representations in speech understanding systems, we argue that it is infeasible to handle the necessary range of concepts using only flat structures. By exploiting the specific nature of the task, we show that it is possible to implement a solution which only slightly extends the representational complexity of the semantic representation language, by permitting an optional single nested level representing a subordinate clause construct. We sketch our solutions to the key problems of producing minimally nested representations using phrase-spotting methods, and writing cleanly structured rule-sets that map temporal and phrasal representations into a canonical interlingual form. Rayner, Bouillon, Santaholma and Nakao 1 Introduction As a subject, automatic speech translation is now a little more than ten years old. First gen- eration systems, like Verbmobil (Wahlster, 2000), Spoken Language Translator (Rayner et al., 2000) and Janus III (Lavie et al., 1997) were essentially proofs of concept. We are now pro- gressing to the stage where people want to build systems that have some claim to be useful: prominent recent examples are NESPOLE! (Lavie et al., 2001), Tongues (Black et al., 2002) and Phraselator (Phraselator, 2004). For obvious reasons, one application area that stands out is medical translation; this paper will focus on representational issues in MedSLT, a medium- vocabulary medical speech translation system (Rayner & Bouillon, 2002; Rayner et al., 2003a). There are many different contexts in which medical translation could potentially be useful. In the scenario targeted by the MedSLT system, we envisage that a doctor wishes to perform a pre- liminary examination of a patient who does not speak the doctor’s language. The system allows the doctor to pose normal examination questions in her own language, translating them into the patient’s language. The …

Content truncated to 500 words