NLPR Translation System for IWSLT 2006 Evaluation Campaign

In this paper we describe a hybrid approach to Chinese-to-English spoken language translation system used for the IWSLT 2006 evaluation campaign. In this system, the phrase-based statistical machine translation (SMT) engine is combined with the template-based machine translation (TBMT) engine and a simple way is proposed to select the best translation from the results generated by the two translation engines. The experiments prove that the combination can improve the performance of translation system. As the input sentences are speech recognition results and have no punctuation information, we restore the punctuation in source sentences in the processing and post-processing. 1