CASIA SMT System For TC-STAR Evaluation Campaign 2006

This paper presents our statistical machine translation (SMT) system for TC-STAR Evaluation Campaign 2006. In our Chinese-to-English SMT system, we use the phrase-based translation model. Considering the characteristics of Chinese-to-English translation, we propose some approaches to improve different tr acing back and hypothesis expansion. For remediation of the intrinsic flaws in statistical framework, we make use of the existing la nguage resources to improve the pr eprocessing model, word alignme nt and the estimation of phrase probability. The preliminary experimental results show that integrating kinds of knowledge resources into the statistical system has the potential to improve th e performance of Chinese-to-English SMT system.