번역학계와 언어서비스업체(LSP)간 산학협력연구: ‘포스트에디팅 생산성’과 ‘기계번역 엔진 성능 비교

Abstract

As machine translation post-editing (MTPE) has become a common practice among language service providers (LSPs), and universities have an urgent need to incorporate MTPE into their curriculum, a common interest on MTPE practices led to the forming of a university-industry joint research team between The Korean Association for Translation Studies (KATS) and The Korea IT Globalization Organization (KIGO). This collaborative study aims to explore the following issues: First, the productivity of MTPE results gauged in terms of two factors, the amount of time spent on post-editing of raw MT outputs against human translation time, and TER (Translation Edit Rate) of MTPE results; Second, error rates of the most representative free online machine translations (FOMT) for the English-Korean language pair, such as Google Translate, Naver’s Papago, and Kakao i in terms of mistranslation, accuracy, terminology, formatting, and spelling. Fifteen undergraduate and graduate students majoring translation participated in the study to produce translation and MTPE results of five IT manuals provided by the participating LSPs. For 13 out of 15 participants, the temporal and technical MTPE efforts were found to be much reduced. Regarding the MT engine error rate, Google Translate was found to have a much lower error rate against Papago and Kakao i in processing IT manuals; however, this result needs further investigation with more variables.