Speech translation for French within the c-STAR II consortium and future perspectives

Despite joining the C-STAR II consortium in late 1996, the CLIPS ++ group succeeded in building the French parts of a multilingual task-oriented spoken dialogue translation system and took part in multilingual, intercontinental demonstrations held on July 22 1999 by CLIPS (France), CMU (United States), ETRI (South Korea), ATR (Japan), IRST (Italy), and UKA (Germany). The challenge was to reach the minimum quality level adequate for handling specific tasks, which is quite higher than what is sufficient for casual chatting and can be achieved by putting together commercially available components. After presenting the modules and the architecture of our C-STAR II demonstrator, we evaluate the results, both externally and internally. While the reactions to the final demonstrations were very positive, and many said that these prototypes should quickly lead to products, we feel that there is still much room for improving the overall quality in significant ways. In the last part, we focus on future avenues of research to further improve the quality of task-oriented speech translation, in particular by defining a more powerful and orthogonal taskoriented semantic pivot, using the linguistic and dialogic context, and generating information usable by speech synthesis to generate better prosody.