ISCA Archive ICSLP 2002
ISCA Archive ICSLP 2002

Improving statistical machine translation for a speech-to-speech translation task

Stephan Vogel, Alicia Tribble

In this paper we investigate if statistical machine translation (SMT) is possible when only a small bilingual corpus is available for training the system. Using additional knowledge sources which are not domain-specific improves the performance of the system considerably. We present results on a speech translation task for German to English. Automatic and human evaluation are used to compare the performance of the SMT system to an interlingua-based translation system.