기계번역 결과물의 오류유형 고찰

Abstract

Advancement in technology leads to rapid development of machine translation. On account of such development, a new way of translating like post-editing is emerging. Post-editing is fixing errors in machine translation output, hence enhancing the quality of machine translation. Effort required by post-editing may vary by genre/type of text and the patterns of errors specific to source text. This pilot study intends to classify machine translation errors in informative text. Firstly, the study provides classification of machine translation errors based on four broad classes: Accuracy, Fluency, Syntax, and Typo. Secondly, errors from English-Korean machine translation of informative texts are analysed with the proposed classification. Lastly, the paper explores occurrence frequency of each error classes and deduces tendencies from the analysis: Incorrect meaning error occurs rather frequently while omission error is found relatively few; Wrong word/phrase order error comes with the incomplete sentence error; Typo errors occur randomly without any patterns. The findings fall short of presenting error patterns due to relatively small size of sample. Future research could look into more predictable error patterns in machine translation that could not be investigated here and might contribute to reducing efforts required by post-editing.