A Statistical Approach to Chinese-to-English Back-Transliteration

A Statistical Approach to Chinese-to-English Back-Transliteration Chun-Jen LEV' 2� Jason S. Telecommunication Labs. Chunghwa Telecom Co., Ltd. Chungli, Taiwan, R.O.C. cjlee@cht.com.tw CHANG2�Jyh-Shing Roger JANG2 2 Department of Computer Science National Tsing Hua University Hsinchu, Taiwan, R.O.C. {jschang, long} @cs.nthu.edu.tw Abstract This paper describes a statistical approach for modeling Chinese-to-English back-transliteration. Unlike previous approaches, the model does not involve the use of either a pronunciation dictionary for converting source words into phonetic symbols or manually assigned phonetic similarity scores between source and target words. The parameters of the proposed model are automatically learned from a bilingual proper name list. The experimental results for back-transliteration indicate that the proposed method provides significant improvement over previous work. 1� Introduction Machine transliteration is very important for research and applications in natural language processing, such as machine translation (MT), cross-language information retrieval (CLIR), and bilingual lexicon construction. Proper nouns are often domain specific and frequently created. It is difficult to handle transliteration using existing bilingual dictionaries. Unfamiliar personal names, place names, and technical terms are especially difficult for human translators to transliterate correctly. In CLIR, the accuracy of transliteration greatly affects the retrieval performance. Recent studies have made great strides toward machine transliteration for many language pairs, such as English/Arabic (Stalls and Knight, 1998; Al-Onaizan and Knight, 2002), English/Chinese (Chen et al., 1998; Wan and Verspoor, 1998; Lin and Chen, 2002), English/Japanese (Knight and Graehl, 1998), and English/Korean (Lee and Choi, 1997; Oh and Choi, 2002). Machine transliteration is classified into two types based on transliteration direction. Transliteration, forward-direction, is the process that converts an original word in the source language into an approximate phonetic equivalent word in the target language, whereas back-transliteration, backward-direction, is the reverse process that converts the transliterated word back into its original word. Most of the previous approaches require a pronunciation dictionary to convert a source word into its corresponding pronunciation sequence. Words with unknown pronunciations may cause problem for transliteration. In addition, using a language-dependent penalty function to measure the similarity between a source word and corresponding transliteration or using handcraft heuristic mapping rules to deal with transliteration may lead to problems when porting to other language pairs. In this paper, we focus on Chinese-to-English back-transliteration. The proposed framework requires no conversion of source words into phonetic symbols. The model is trained automatically on a bilingual proper name list. The remainder of the paper is organized as follows: Section 2 presents the proposed statistical transliteration model (STM) and describes the model parameters. In Section 3, we describe the framework to deal with back-transliteration. Experimental setup and the results of the evaluation are presented in Section 4. Concluding remarks are made in Section 5. 2� Statistical Transliteration Model One can consider machine transliteration as a noisy channel. Under the noisy channel model, the back-transliteration problem is to find the most probable word E from the given transliteration C. Let P(E) be the probability of a word E, then, for a given transliteration C, the back-transliteration 310 probability of a word …

Content truncated to 500 words