The quality and productivity of post-edited machine translation in comparison with human translation: A case of Korean-English legal translation

This paper addresses the quality and productivity issues in post-edited machine translation, analyzing Korean-English legal translation outputs produced in two ways: human translation and post-edited machine translation. For this study, two sets of translated texts were produced by professional legal translators and student translators. The quality of the raw machine translation by DeepL was assessed through automatic and manual evaluations, and its scores were compared with those of human and post-edited machine translation outputs. To investigate the productivity of post-editing, we analyzed the temporal and technical efforts involved in human translation as well as post-editing. The analysis indicated the superiority of post-edited machine translation both in terms of quality and productivity. Compared with human translation, post-editing improved the average number of processed words by about 20%, while significantly reducing technical efforts. Correction rates demonstrated that the major errors, which are related to accuracy, fluency, and terminology, in the machine translation output were effectively addressed through post-editing. Despite the limitations of this small-scale study, the findings suggest that post-editing has the potential to efficiently elevate the machine translation output quality to a level that surpasses that of human translation, and thus facilitates translation of legal texts of medium difficulty.