인간번역과 기계번역 및 포스트에디팅의 오류 비교: 한영 법률 번역 사례를 중심으로

This paper addresses the quality of machine translation (MT) and post-editing (MTPE) outputs in comparison with human translation (HT) in the context of Korean-English legal translation, focusing on error analysis. Using the three major criteria of accuracy, fluency, and terminology, the study examines errors in twelve from-scratch translations and twelve post-edited MT outputs as well as the MT output generated by DeepL Translator. Error comments provided by three human evaluators were analyzed to identify error types and occurrences in the three different output types. Our findings reveal that MTPE generated the fewest errors, while MT exhibited the highest error count. Across all output types (HT, MT, and MTPE), accuracy errors were the most prevalent, with distortion being the predominant type. Fluency errors ranked next in frequency for HT and MTPE, while terminology errors were more common in MT than in HT and MTPE. Further analysis indicates that fluency and terminology errors were generally corrected in MTPE, but accuracy errors, particularly distortion, requiring substantial cognitive effort, often remained uncorrected. In terms of prevalent error types, MTPE demonstrated a closer resemblance to HT than MT. These findings underscore the potential of MTPE to yield high-quality legal text translations with reduced errors. However, HT remained advantageous in terms of producing accurate translations, particularly with regard to meaning transfer. This study contributes valuable insights to the ongoing discourse on the efficacy of different translation methods in the legal domain.