한국어-중국어 신경망 기계 번역에 대한 수동평가: 사회과학과 기술과학 카테고리에서 나타나는 구글 번역(GNMT)과 파파고(N2MT)의 정확도 (Evaluation of Korean-Chinese Neural Network Machine Translation-Differences between Google Translation (GNMT) and Papago (N2MT) in the Fields of Social Science and Technical Science)

서지정보

· 발행기관 : 한국외국어대학교 통번역연구소
· 수록지 정보 : 통번역학연구 / 24권 / 4호 / 175 ~ 206페이지
· 저자명 : 황지연, 양슬아

초록

This study conducted machine translation evaluation using GNMT and N2MT machine translation programs. The subject area is articles in the fields of social science and technical science. The evaluation items at the level of words, phrases, sentences, and text were presented, and the accuracy and fluency of Korean-Chinese machine translation were evaluated and analyzed in a 10-point scale. This paper categorizes the types of errors repeatedly appearing in machine translation into four and applied them as top evaluation items. Evaluation models were presented by supplementing and revising several evaluation models presented in previous studies. We will apply this to the text to be analyzed and examine the accuracy and difference of GNMT and N2MT. The analysis results are as follows. First, as a result of field evaluation, GNMT translated social science and technical science with almost the same accuracy. On the other hand, N2MT was significantly less accurate in translating technical science than in translating social science. Second, there were differences among error types, which were observed in translations of both GNMT and N2MT. The highest number of errors occurred at the word level for both neural network translators.

영어초록

This study conducted machine translation evaluation using GNMT and N2MT machine translation programs. The subject area is articles in the fields of social science and technical science. The evaluation items at the level of words, phrases, sentences, and text were presented, and the accuracy and fluency of Korean-Chinese machine translation were evaluated and analyzed in a 10-point scale. This paper categorizes the types of errors repeatedly appearing in machine translation into four and applied them as top evaluation items. Evaluation models were presented by supplementing and revising several evaluation models presented in previous studies. We will apply this to the text to be analyzed and examine the accuracy and difference of GNMT and N2MT. The analysis results are as follows. First, as a result of field evaluation, GNMT translated social science and technical science with almost the same accuracy. On the other hand, N2MT was significantly less accurate in translating technical science than in translating social science. Second, there were differences among error types, which were observed in translations of both GNMT and N2MT. The highest number of errors occurred at the word level for both neural network translators.

참고자료

· 없음