《MTS-2016-Xiao》.pdfVIP

Document-level Consistency Verification in Machine Translation Tong Xiao , Jingbo Zhu , Shujie Yao , Hao Zhang †Natural Language Processing Lab, Northeastern University ‡Key Laboratory of Medical Image Computing, Ministry of Education {xiaotong,zhujingbo}@ {yaosj, zhanghao}@ tant in certain applications such as translation of Abstract legal documents and government documents. In some cases, consistency is even regarded as one of Translation consistency is an important issue in the primary quality measurements of translation document-level translation. However, the con- (He et al., 2009). sistency in Machine Translation (MT) output is However, directly modeling the translation generally overlooked in most MT systems due to problem on the whole document is a challenging the lack of the use of document contexts. To ad- issue due to its high complexity. It is even intract- dress this issue, we present a simple and effec- tive approach that incorporates document con- able to implement or run such a translation system texts into an existing Statistical Machine Trans- in practice. To ease the problem, a general solution lation (SMT) system for document-level transla- is to view the source text as a series of independent tion. Experimental results sh