Lexical Diversity in Statistical and Neural Machine Translation

Neural machine translation systems have revolutionized processes in terms of quantity and speed recent years, they even been claimed to achieve human parity. However, the quality their output has also raised serious doubts concerns, such as loss lexical variation, evidence “machine translationese”, its effect on post-editing, which results “post-editese”. In this study, we analyze outputs three English Slovenian diversity different genres. Using both quantitative qualitative methods, one statistical two neural systems, compare them a reference translation. Our analyses based metrics show diverging results; however, particularly ones, mostly exhibit larger than counterparts. Nevertheless, method shows that these are not always reliable tool assess true lot “creativity”, especially by is often unreliable, inconsistent, misguided.