The Gap between NMT and Professional Translation from the Perspective of Discourse
Ying Hua Qin, Jie Zhang, Xiaoping Lü · 2019
Neural Machine Translation (NMT) attracts an intensification of research efforts in recent years. The claim that the quality of machine translation is matching human translator requires extraordinary evidences. It is almost certain that machine translation so far cannot replace or exceed human translator, but what is the gap between NMT and professional translator? We truly investigate the quality difference between the well-known Google Neural Machine Translation (GNMT) and professional translators at the document level. We build an evaluation model based on the discourse cohesion theory to compare the usage of grammatical and semantical devices in Chinese-to-English translations including reference, substitution, ellipsis, conjunction and lexical cohesion. The evaluation results show that there is a significant gap between NMT and professional translator in document translations. In a comprehensive point of view, for the professional translator, translation is a kind of arts; while for the computers, it is still a kind of technologies.