Abstract
In Koga (1999), I evaluated an Korean-Japanese machine translation software by categorizing mistranslations into 4 groups: mistranslation of words themselves, mistranstation of co-occurrence, grammatical mistranslation and textual mistranslation. According to the result, more than half of the mistranslations proved to be that of cooccurrence. In this paper, I further analyzed the mistranslations of co-occurrence by subcategorizing them into 2 groups: co-occurrence of words and grammatical co-occurrence, and by checking the grammatical feature of words or grammatical morphemes of the co-occurrence. As a result, it became clear that these mistranslations are very restricted in these subcategories.
