TREATING 'HATA'-CONSTRUCTION IN KOREAN-CHINESE MT

摘要

This paper addresses the treatment of light-verb construction with 'hata' in Korean-Chinese machine translation. The 'hata' (= to do) verb, which is frequently used in Korean, can be syntactically realized with two different forms representing, however, the same propositional meaning: 'X-hata' (= to do X) and 'X-lul hata' (= to do X). Based on this observation, we propose two verb constructions share the verb patterns, in other words, they share the subcategorization frames. This kind of sharing of information can eliminate the redundancy of verb patterns and increase the coverage of 'X-hata' patterns. However, it could pose a generation problem, since a verb pattern contains not only a source language pattern part for the analysis but also a corresponding target language pattern part for the generation. To solve this, we classify 'X-lul hata' construction according to its corresponding Chinese expression, and propose the generation process be tailored for each type. The evaluation of the proposed method shows 90.6% success rate.