Analysis of Korean-to-English Machine Translation Systems’ Treatment of Passives
Chae-Eun Kim · Journal of Pan-Pacific Association of Applied Linguistics · 2022
This study explores how Korean-to-English machine translation (MT) systems (e.g., Google Translator, NAVER Papago) deal with Korean passive structures. Cross-linguistically, Korean and English passives show different ways to construct passive-voice sentences from active structure. English passives including with be+p.p. may have a ‘by phrase’ after the verb, which perform the agent of the action. However, Korean has three different passive structures and a rich repertoire of affixes, including case markers, which play critical roles in the production and comprehension of passives, complicating translation to English. Moreover, the grammatical constructions that express passive voice in Korean and English differ in terms of frequency in both written and spoken contexts. To explore production accuracy by two MTs, 213 Korean passive sentences were translated into English passives. This finding shows that current Korean-to-English MTs almost always produce translations that are correct in meaning when they generally translate Korean passives into English passives. They also translate Korean passives into English active voice constructions at significant rates.