Chinese-English Backward Transliteration Assisted with Mining Monolingual Web Pages

Proceedings of ACL-08: HLT, pages 541–549, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics Chinese-English Backward Transliteration Assisted with Mining Mono- lingual Web Pages Fan Yang, Jun Zhao, Bo Zou, Kang Liu, Feifan Liu National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China {fyang,jzhao,bzou,kliu,ffliu}@nlpr.ia.ac.cn Abstract In this paper, we present a novel backward transliteration approach which can further as- sist the existing statistical model by mining monolingual web resources. Firstly, we em- ploy the syllable-based search to revise the transliteration candidates from the statistical model. By mapping all of them into existing words, we can filter or correct some pseudo candidates and improve the overall recall. Secondly, an AdaBoost model is used to re- rank the revised candidates based on the in- formation extracted from monolingual web pages. To get a better precision during the re- ranking process, a variety of web-based in- formation is exploited to adjust the ranking score, so that some candidates which are less possible to be transliteration names will be as- signed with lower ranks. The experimental re- sults show that the proposed framework can significantly outperform the baseline translit- eration system in both precision and recall. 1 Introduction* The task of Name Entity (NE) translation is to translate a name entity from source language to target language, which plays an important role in machine translation and cross-language informa- tion retrieval (CLIR). Transliteration is a subtask in NE translation, which translates NEs based on the phonetic similarity. In NE translation, most person names are transliterated, and some parts of location names or organization names also need to be trans- literated. Transliteration has two directions: for- ward transliteration which transforms an original name into target language, and backward translit- eration which recovers a name back to its original expression. For instance, the original English per- *Contact: Jun ZHAO, jzhao@nlpr.ia.ac.cn. son name “Clinton” can be forward transliterated to its Chinese expression “克/ke 林/lin顿/dun” and the backward transliteration is the inverse process- ing. In this paper, we focus on backward translit- eration from Chinese to English. Many previous researches have tried to build a transliteration model using statistical approach [Knight and Graehl, 1998; Lin and Chen, 2002; Virga and Khudanpur, 2003; Gao, 2004]. There are two main challenges in statistical backward trans- literation: First, statistical transliteration approach selects the most probable translations based on the knowledge learned from the training data. This approach, however, does not work well when there are multiple standards [Gao, 2004]. Second, back- ward transliteration is more challenging than for- ward transliteration as it is required to disambiguate the noises introduced in the forward transliteration and estimate the original name as close as possible [Lin and Chen, 2002]. One of the most important causes in introducing noises is that: some silent syllables in original names have been missing when they are transliterated to target lan- guage. For example, when “Campbell” is translit- erated into “坎/kan贝/bei尔/er”, the “p” is missing. In order to make up the disadvantages …

Content truncated to 500 words