With the recent popularity of free online machine translation services, researchers have been interested in the evaluation of machine translation systems. The increasing literature in the field, however, has not answered two important questions: how accurate the online machine translation for English to Korean is; and how the translation quality is determined by linguistic factors like argument structure and transitivity alternations. The purpose of this paper is to investigate linguistic determinants for successful machine translation of English into Korean, and to suggest possible directions for improving the translation quality from the perspective of lexical semantics. For this, we tested the accuracy of Google Translate (=GT) for English to Korean by using 260 sentences produced by Levin``s (1993) English verb classification. The quality of the translation was assessed with reference to Hampshire & Salvia``s (2010) criteria of clarity and fidelity. Results from the cumulative logit multinomial regression analysis showed that the lexical semantic differences between English and Korean caused the failure of translation, and that GT could not translate the conative and middle intransitive constructions in English accurately. In the end, we argue that the quality of machine translation can be improved above traditional non-linguistic approaches by incorporating in-depth linguistic principles.