Robust spoken translation at ITC-IRST

MT Summit VII Sept.1999 Robust Spoken Translation at ITC-IRST Gianni Lazzari ITC-IRST Trento, Italy Abstract In this paper the ITC-irst research issues and approach to the spoken translation problem will be presented together with a description of the demonstration system developed in the framework of C-STAR II Consortium. The challenge of future applications in the e- commerce and e-service sectors will also be presented and discussed. 1 Introduction Significant progress has been made in the field of human language technologies. Various tasks like con- tinuous speech recognition for large vocabulary, speaker and language identification, spoken informa- tion inquiry in restricted domains are today feasible and different prototypes and systems are running. The spoken translation problem on the other hand is still a significant challenge[16]. Nevertheless recently im- portant advances have been demonstrated approaching this problem as a human to human communication task carried on in a restricted domain. In this case bi- directional, real time operation is necessary, but fairly low quality is acceptable when communication is achieved. A Spoken Translation System does not need to give a complete correct solution, if it produces a sufficient expression in the target language satisfying the human dialog situation. If the human to human communication is also supported by multimedia in- formation this approach is even more convincing. This solution to the problem of spoken translation is also suggested by the needs of the information society. The diffusion of Internet and the related services is asking for high sophisticated tools for human to hu- man communication over the Net. Web Phone. Video Call Centers, Call Centers, Videoconferencing Sys- tems allow today the possibility of a global communi- cation infrastructure, despite the lack of adequate band. This infrastructure offers high human to human communication capabilities with the only exception of the language, which is and will be for the future a sig- nificant barrier to overcome. ITC-irst has been working on spoken language tech- nologies since 1987 and on spoken translation since 1995, when joined the C-STAR II consortium as a partner. The approach chosen and pursued at ITC-irst to the spoken translation problem is based on the fol- lowing assumptions: • optimize the capability to transmit the meaning of the communication focusing on the robustness of the system intended as capability to manage spontaneous speech and ill formed language (the so called ungram- matical spoken language) and on aid of multimedia in- formation. • focus on the scalability and portability of the sys- tern. This means to be able to extend a given domain and/or to port the system to different domains with rea- sonable effort. • demonstrate the usefulness of the use of 'interlin- gua' or 'interchange format' approaches. Considering the translation task as a coupling of two different passes, an analysis step which produces a suitable interlingua from a given speech signal, a gen- eration step, which produces a spoken sentence in the target language starting from an interlingua represen- tation. In this paper first of all the ITC-irst research issues …

Content truncated to 500 words