Second language processing of errors in Korean-to-English machine-translated output

Abstract

While previous investigations on online machine translation (MT) in language learning have analyzed how second language (L2) learners use and post-edit MT output, no study as of yet has investigated how the learners process MT errors and what factors affect this process using response and reading times. The present study thus investigates L2 processing of MT errors that are caused by syntactic, morphological, and semantic differences between the source and target language and also examines how L2 proficiency and visual display affect this process. Forty-seven Korean learners of English participated in an acceptability judgment task in which they read a Korean sentence and then its translated counterpart in English and had to judge the accuracy of the translated sentence on a four-point scale. The response latencies for the accuracy judgment as well as the total reading times of source and target sentences were measured. The results revealed that (a) learners generally find it harder to reject mistranslations than to accept correct translations, (b) high and low proficiency learners focus on different aspects of language when processing translated output, and (c) constant visual access to the source text does not facilitate but rather interferes with processing MT errors.