Adding Statistical Machine Translation Adaptation to Computer-Assisted Translation

Statistical machine translation (SMT) has proven effective for general purpose language translation but not for highly specialized domains like medicine, military operations, and law enforcement, which have their own technical jargon. We present a novel approach for iteratively incorporating a human translator in the loop to adapt SMT models to a particular domain. We show how these models can be made accessible via Web services and integrated with computer-assisted translation (CAT) tools. In this report, we describe a novel human-in-the-loop post-editing domain adaptation algorithm for refining SMT models using the Joshua decoder and integrate it with a CAT tool called OmegaT.