ETRI-Knowledge Sharing Plaform



논문 검색
구분 SCI
연도 ~ 키워드


학술지 영한 번역 메모리의 구조화 연구
Cited - time in scopus Download 3 time Share share facebook twitter linkedin kakaostory
최승권, 김영길
번역학연구, v.10 no.3, pp.193-216
09MS5200, 한중영 대화체 및 기업문서 자동번역 기술개발, 김영길
This paper aims at developing a structured translation memory TM+ which can resolve the coverage problem of an string-based translation memory TM and enhance a translation quality of English-Korean machine translation system.The existing TM is basically a type of bilingual corpus with full-form words, while TM+ including TM consists of different translation memories made by the following steps: 1) pre-processing, 2) deletion of sentence-initial adverbs and expansion, 3) chunking of proper noun and substitution of it by a variable PRN, 4) chunking of numeral expression and substitution of it by a variable NUM, 5) chunking of base noun phrase and substitution of it by a variable BNP, and 6) chunking of idioms and substitution of the remainders by their corresponding variables.We evaluated 200 test sentences to compare the translation accuracy of machine translation system with TM with one of machine translation system with TM+. The experimental result shows that while the translation accuracy of machine translation system with TM is 74.44%, the translation accuracy of TM+ amounts 82.40%. From this result we could know that TM+ rose a translation quality of 7.96%.In the near future we have plans to segment the steps of TM+ in more detail, introduce the semantic information for numeral expression, and develop the alignment technology between source segments and their target equivalents. We hope that TM+ be applicable to the computer-aided translation tools like TRADOS and allow the professional translators to translate easier by using it.
KSP 제안 키워드
Bilingual corpus, Computer-aided Translation, Experimental Result, Machine Translation(MT), Noun phrase, Pre-processing, Translation Accuracy, Translation Memory(TM), Translation quality, Translation system, coverage problem