AMachine Learning Method to Distinguish Machine Translation from Human Translation

This paper introduces a machine learning ap-proach to distinguish machine translation texts from human texts in the sentence level au-tomatically. In stead of traditional methods, we extract some linguistic features only from the target language side to train the predic-tion model and these features are independent of the source language. Our prediction mod-el presents an indicator to measure how much a sentence generated by a machine translation system looks like a real human translation. Furthermore, the indicator can directly and ef-fectively enhance statistical machine transla-tion systems, which can be proved as BLEU score improvements.