문학번역 적용을 위한 기계번역의 현주소

초록

Along with the development of Neural Machine Translation (NMT), it has been predicted that machine translation can be used for various text types. However, predictions without objectives and scientific data are unreliable and even misleading. Hence, this paper tries to present an objective assessment of the current state of NMT and to investigate whether NMT can be applicable to Korean-to-English literary translation. The starting point is to step back from conventional mistranslation-centric approaches and to accommodate the findings both from natural language processing research and translation studies, since machine translation research is multidisciplinary in its nature. To that end, this paper introduces the structural limitations of NMT engines which have already been reported by engineering scholars, and tries to prove the limitations which cause distortion of the original meaning and literary effect. To build an objective data set, the machine translation outcomes from two free on-line translation engines were used to translate four Korean novels. Three of them were compared against human translation, segment by segment, to underline the difference and shortcomings of machine translation. In addition, machine translation outcomes of one novel, without matching human translation, were used just for data validation. Among the many constraints of NMT, this paper selected and focused on unpredictable omission, difficulty in reading the context, long sentence processing, and out-of-vocabulary processing. After reviewing all these drawbacks, this paper makes some suggestions as to how we should view machine translation as a potential tool for future literary translation.

키워드