"Speech-to-Speech Translation: The Project Verbmobil". In: Y. Kosarev (ed.), Proceedings of SPECOM96. St. Petersburg (with Jan W. Amtrup)


The research project Verbmobil combines the research branches of continuous speech recognition and machine translation. A consortium of more than 20 universities and 8 companies in Germany as well as two partner institutes in USA and Japan is enganged in the development of a system for translation of spontaneous and continuous speech input to speech output. In the first phase the language pair German and Japanese have been chosen with English as a presentation language (to check the translation quality). The domain of discourse is appointment scheduling. The final application may be an appointment negotiation among two industrial managers from Japan and Germany. In the second project phase the domain will be extended to include travel planning discussions. There exists a prototype implementation integrating the modules of the partners. Additionally background research has been performed to investigate prablems e.g. of dialogue and discourse modelling, stochastic parsing and innovative software architectures. This paper gives an overview of the project, its goals and results. It describes in more details research done on integrative architectures. This paper (except chapter 4) is mostly based on official Verbmobil documents (see Wahlster and Kay)

To download the paper click the title, please!

You will receive

    file format: rtf
    encryption: gzip
    size: 26k

In case of any problems please send a mail to vhahn@informatik.uni-hamburg.de
  1. Summary