Integrating Translation Memory into Phrase-Based Machine Translation during Decoding

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 11–21, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Integrating Translation Memory into Phrase-Based Machine Translation during Decoding Kun Wang† Chengqing Zong† Keh-Yih Su‡ †National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China ‡Behavior Design Corporation, Taiwan †{kunwang, cqzong}@nlpr.ia.ac.cn, ‡kysu@bdc.com.tw Abstract Since statistical machine translation (SMT) and translation memory (TM) complement each other in matched and unmatched regions, integrated models are proposed in this paper to incorporate TM information into phrase-based SMT. Unlike previous multi-stage pipeline approaches, which directly merge TM result into the final output, the proposed models refer to the corresponding TM information associat- ed with each phrase at SMT decoding. On a Chinese–English TM database, our experi- ments show that the proposed integrated Mod- el-III is significantly better than either the SMT or the TM systems when the fuzzy match score is above 0.4. Furthermore, integrated Model-III achieves overall 3.48 BLEU points improvement and 2.62 TER points reduction in comparison with the pure SMT system. Be- sides, the proposed models also outperform previous approaches significantly. 1 Introduction Statistical machine translation (SMT), especially the phrase-based model (Koehn et al., 2003), has developed very fast in the last decade. For cer- tain language pairs and special applications, SMT output has reached an acceptable level, es- pecially in the domains where abundant parallel corpora are available (He et al., 2010). However, SMT is rarely applied to professional translation because its output quality is still far from satis- factory. Especially, there is no guarantee that a SMT system can produce translations in a con- sistent manner (Ma et al., 2011). In contrast, translation memory (TM), which uses the most similar translation sentence (usual- ly above a certain fuzzy match threshold) in the database as the reference for post-editing, has been widely adopted in professional translation field for many years (Lagoudaki, 2006). TM is very useful for repetitive material such as updat- ed product manuals, and can give high quality and consistent translations when the similarity of fuzzy match is high. Therefore, professional translators trust TM much more than SMT. However, high-similarity fuzzy matches are available unless the material is very repetitive. In general, for those matched segments1, TM provides more reliable results than SMT does. One reason is that the results of TM have been revised by human according to the global context, but SMT only utilizes local context. However, for those unmatched segments, SMT is more re- liable. Since TM and SMT complement each other in those matched and unmatched segments, the output quality is expected to be raised signif- icantly if they can be combined to supplement each other. In recent years, some previous works have in- corporated TM matched segments into SMT in a pipelined manner (Koehn and Senellart, 2010; Zhechev and van Genabith, 2010; He et al., 2011; Ma et al., 2011). All these pipeline approaches translate the sentence in two stages. They first determine whether the …

Content truncated to 500 words