Representation and Recognition Method for Multi-Word Translation Units in Korean-to-Japanese MT System

Representation and Recognition Method for Multi-Word Translation Units in Korean-to-Japanese MT System Kyonghi Moon Dept. of Computer Science & Engineering Pohang Univ. of Science and Technology San 31 Hyoja-dong Nam-gu, Pohang 790-784 Republic of Korea khmoon@kle.postech.ac.kr Jong-Hyeok Lee Dept. of Computer Science & Engineering Pohang Univ. of Science and Technology San 31 Hyoja-dong Nam-gu, Pohang 790-784 Republic of Korea jhlee@postech.ac.kr Abstract Due to grammatical similarities, even a one-to-one mapping between Korean and Japanese words (or morphemes) can usually result in a high quality Korean-to-Japanese machine translation. However, multi-word translation units (MWTU) such as idioms, compound words, etc., need an n-to-m mapping, and their component words often do not appear adjacently, resulting in a discontinuous MWTU. During translation, the MWTU should be treated as one lexical item rather than a phrase. In this paper, we define the types of MWTUs and propose their representation and recognition method depending on their characteristics in Korean-to-Japanese MT system. In an experimental evaluation, the proposed method turned out to be very effective in handling MWTUs, showing an average recognition accuracy of 98.4% and a fast recognition time. 1 Introduction As a transfer problem in a machine translation (MT), lexical and structural differences exist between source and target languages, which requires 1-n, m-n, or n-1 mapping strategies for machine translation system. For such mapping strategies, we need to treat several (n, or m) words (or morphemes) as a single translation unit. Although some researches (D.Santos,1990; Linden E.,1990; Yoon Sung Hee, 1992; Ha Gyu Lee, 1994; D.Arnold,1994) employ the term “idiom” for these units, we prefer MWTU (Multi-Word Translation Unit) because it is a more general and broader term for MT environment. Up to now, some research has focused on recognition and transfer of MWTUs, although very little research has been undertaken for Korean-to-Japanese machine translation systems (Seon-Ho Kim,1997). In previous researches, some tended to simplify the problem by treating only special types of MWTUs, while others had some recognition errors and took too much recognition time because they did not restrict the recognition scope (D.Santos,1990; Yoon Sung Hee,1992; Ha Gyu Lee, 1994; Seon-Ho Kim,1997). For a Korean-to-English MT, Lee and Kim (Ha Gyu Lee,1994) uses only weak restrictions like adjacent information for recognition scope. However, their method needs stronger restrictions to resolve recognition errors and to speed up the process. Although some differences exist depending on which kinds of source and target languages are dealt with, MWTUs in Korean-to-Japanese MT frequently have their component words close together, so that one can predict the location of their separated component words. For this reason, we can enhance the recognition accuracy and time effectively by restricting the recognition scope according to the characteristics of an MWTU rather than taking the whole sentence as the scope. Moreover, the method by Lee and Kim (Ha Gyu Lee,1994) deals with only surface-level consistency without considering word order because Korean has almost free word order. It is obvious that the method can deal with variable word-order MWTUs, but some incorrect recognition results are …

Content truncated to 500 words