Using Bilingual Semantic Information in Chinese-Korean Word Alignment

USING BILINGUAL SEMANTIC INFORMATION IN CHINESE- KOREAN WORD ALIGNMENT Jin-Xia Huang"' , Key-Sun Choi* *KORTERM, Computer Science Division Korean Advanced Institute of Science and Technology 373-1 Yusong-ku Taejon 305-701 Korea **Yanbian University of Science and Technique, Beishan Street, Yakji, 133001, PR.China Email: {hglaschoi }(& world. kaist. ac.kr ABSTRACT This pape clarifies the definition of alignment from the viewpoint of linguistic similarity. We propose new method for the alignment betwee the languages that do not belong to the same language family. On the contrary to most of the previously proposed methods that rely heavily on statistics, our method attempts to use linguistic knowledge to overcome the problems of statistical model. 1. INTRODUCTION 1.1 Previous Works Bilingual corpus provides more information than monolingual one (Dagan, 1991). In recent years many works have been done on word alignment after the research on section, paragraph, sentence, and phrase level alignment. Word alignment works not only for th automatic construction of bilingual dictionaries and other useful resourc s, but also for the various applications such as machine translation (MT) and word sense disambiguation. Statistical approach has been used as main technique in most alignment systems (Gale, 1991; Brown, 1993; Dagan, Church, Gale, 1993). Correlation information of bilingual lexicon is mainly employed in statistical approach, and other information, such as character information or position information is used additionally (Brown, 1993; Dagan, 1993). In the alignment of some language pairs that do not belong to the same language family, word list (Wu, 1994) or functional word information (Shin and Choi, 1996) was employed additionally. In contrast to the systems that mainly rely on statistical approach, the English-Chinese alignment system developed by Ker (1997) used a class-based algorithm in its word alignment, and the syste did not use any statistical technique. The result show ed that this new approach could overcome the lower coverage of the statistical approach whil gaining high precision, even if his system was only based on small test set. Ker's work clews us the feasibility of pure linguistic approach in the resol utio of the alignment problem. Phrase level alignment is one of the most difficult problems in word alignment. Some statistical approaches have been used to resolve the problem, but most of them get hold of not linguistic phrases but statistical ones (Wu, 1994; Shin & Choi, 1996). Using parser maybe a solution to this problem (Turcato, 1998), but it can cause other problems such as data sparseness because of the syntax diversity between th two languages. For the improvement of correction, some research restrained the alignment object to som special phrase such as compound noun (Kupiec, 1993; Kumano, 1994). The research described in this paper is sponsored by Advanced Information Technology Research Center (AITrc) and Korean Science and Engineering Foundation (KOSEF) Phrase level alignment is especially much more difficult when the language pairs do not belong to th same language family, becaus the syntax structure and the lexical formation are much more different in this case there are more cases out of …

Content truncated to 500 words