Chinese-English SMT for cross-language dialogue agent support

Advances in Statistical Machine Translation (SMT) for breaking language barrier have been seen in recent years, and there is huge demand on cross-language dialog communication between people. In this paper we propose to leverage SMT for supporting cross-language dialog communication. Several techniques are applied to improve the performance on a dialog domain, including rescoring, system combination, post-editing etc. Furthermore, we develop a SMT system targeting dialog management. Various kinds of features specific to dialog modeling, such as reiteration, synonyms in one dialog, are proposed to capture the coherent property of dialog translation. We use travel domain Chinese-English dialog corpus to study the proposed method. Experiments show that the proposed machine translation system provides better translation quality on dialog translation than common phrase-based machine translation system.