A Semantic-Specific Model for Chinese Named Entity Translation

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 138–146, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP A Semantic-Specific Model for Chinese Named Entity Translation Yufeng Chen and Chengqing Zong National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences Beijing, China, 100190 {chenyf,cqzong}@nlpr.ia.ac.cn Abstract We observe that (1) it is difficult to combine transliteration and meaning translation when transforming named entities (NE); and (2) there are different translation variations in NE translation, due to different semantic informa- tion. From this basis, we propose a novel se- mantic-specific NE translation model, which automatically incorporates the global context from corpus in order to capture substantial semantic information. The presented approach is inspired by example-based translation and realized by log-linear models, integrating mo- nolingual context similarity model, bilingual context similarity model, and mixed language model. The experiments show that the seman- tic-specific model has substantially and con- sistently outperformed the baselines and re- lated NE translation systems. 1 Introduction Named entity (NE) translation, which transforms a name entity from source language to target lan- guage, plays a very important role in trans- lingual language processing tasks, such as ma- chine translation and cross-lingual information retrieval. Generally, NE translation1 (1) The combination of transliteration and meaning translation. Either transliteration or meaning translation is only a subtask of NE translation. There has been less work devoted to includes translitera- tion and meaning translation. Recently, many researches have been devoted to NE translitera- tion (most person names) or NE meaning transla- tion (organization names) individually. However, there are still two main challenges in statistical Chinese-English (C2E) NE translation. 1 NE translation referred to in this paper denotes bilingual NE transformation (either transliteration or meaning transla- tion), and meaning translation is proposed as distinct from transliteration. the combination of transliteration and meaning translation for translating NEs. (2) The selection of NE translation variations. Segments in different NEs could be translated differently due to NEs’ origins and enrich lan- guage phenomenon (Huang et al., 2005). As shown in Table 1, the same Chinese character “金” is translated into different English varia- tions (highlighted in aligned parts). Transliteration variations 金炳华 —— Jin Binghua 金成勋 —— Kim Sung-Hoon 何塞 华金 布伦纳 —— Jose Joaquin Brunner 若阿金·希潘德 —— Joaquim Chipande 马丁路德金 —— Martin Luther King 金丸信 —— Kanemaru Shin 米斯金 —— Miskine 麦金托什 —— Aaron Mcintosh 文森特·伯金 —— Vincent Burgen 埃尔金·杰拉辛 —— Ergin Celasin 阿利亚夫金 —— Alyavdin 卡列伊金 —— Kaleikin …… Meaning translation variations 阿斯特基金 —— Astor Fund 北京冶金学院 —— Beijing Institute of Metallurgy …… Table 1．C2E Translation variations of a charac- ter “金” in different instances Furthermore, we randomly extract 100 Chi- nese characters from the person names of LDC2005T34 corpus, and find out all the charac- ters have more than one translation variations. And each character has about average 7.8 trans- lation variations. Also, (Li et al., 2004) have in- dicated that there is much confusion in C2E transliteration and Chinese NEs have much lower perplexity than English NEs. According …

Content truncated to 500 words