Korean and Japanese are very similar in their linguistic structures. Therefore, one may expect quite a high accuracy of Korean-Japanese machine translation compared to other machine translation from Japanese to some language structurally very different from Japanese, for example, like English. However, using one of the K-J machine translation software available on the market, one ends up finding the result far short of his expectation. In this paper, linguistic mistranslations are categorized into four groups: mistranslation of words themselves, mistranslation of co-occurrence, grammatical mistranslation and textual mistranslation. By translating several texts using one of such software, we examined the number of the mistranslations of each category. According to the result, more than half of the mistranslations proved to be that of co-occurrence.
View full abstract