Name-aware Machine Translation

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 604–614, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Name-aware Machine Translation Haibo Li† Jing Zheng‡ Heng Ji† Qi Li† Wen Wang‡ † Computer Science Department and Linguistics Department Queens College and Graduate Center, City University of New York New York, NY, USA 10016 {lihaibo.c, hengjicuny, liqiearth}@gmail.com ‡ Speech Technology & Research Laboratory SRI International Menlo Park, CA, USA 94025 {zj, wwang}@speech.sri.com Abstract We propose a Name-aware Machine Translation (MT) approach which can tightly integrate name processing into MT model, by jointly annotating parallel cor- pora, extracting name-aware translation grammar and rules, adding name phrase table and name translation driven decod- ing. Additionally, we also propose a new MT metric to appropriately evaluate the translation quality of informative words, by assigning different weights to differ- ent words according to their importance values in a document. Experiments on Chinese-English translation demonstrated the effectiveness of our approach on en- hancing the quality of overall translation, name translation and word alignment over a high-quality MT baseline1. 1 Introduction A shrinking fraction of the world’s Web pages are written in English, therefore the ability to access pages across a range of languages is becoming in- creasingly important. This need can be addressed in part by cross-lingual information access tasks such as entity linking (McNamee et al., 2011; Cas- sidy et al., 2012), event extraction (Hakkani-Tur et al., 2007), slot filling (Snover et al., 2011) and question answering (Parton et al., 2009; Parton and McKeown, 2010). A key bottleneck of high- quality cross-lingual information access lies in the performance of Machine Translation (MT). Tradi- tional MT approaches focus on the fluency and accuracy of the overall translation but fall short in their ability to translate certain content word- s including critical information, especially names. 1Some of the resources and open source programs devel- oped in this work are made freely available for research pur- pose at http://nlp.cs.qc.cuny.edu/NAMT.tgz A typical statistical MT system can only trans- late 60% person names correctly (Ji et al., 2009). Incorrect segmentation and translation of names which often carry central meanings of a sentence can also yield incorrect translation of long con- texts. Names have been largely neglected in the prior MT research due to the following reasons: • The current dominant automatic MT scoring metrics (such as Bilingual Evaluation Under- study (BLEU) (Papineni et al., 2002)) treat all words equally, but names have relative low frequency in text (about 6% in newswire and only 3% in web documents) and thus are vast- ly outnumbered by function words and com- mon nouns, etc.. • Name translations pose a greater complexity because the set of names is open and highly dynamic. It is also important to acknowledge that there are many fundamental differences between the translation of names and other tokens, depending on whether a name is ren- dered phonetically, semantically, or a mixture of both (Ji et al., 2009). • The artificial settings of assigning low …

Content truncated to 500 words