Study of Machine Translated Segment Pairs

A new paradigm requires a leap of faith. Download this free report of a study that evaluated of 73,150 translated segment pairs, representative of 31 professional translators’ daily translation work. The study covers 16 unique language pairs with 5 source languages and 13 target languages across 13 different subject domains. It explores new benchmarks that an individual translator can use to predict his or her own experience when using a machine translation (MT) system as an assistive translation technology to perform daily work. It also analyzes the differences between Google’s neural MT (GNMT) and customized statistical MT (CSMT) relative to the traditional “Promised Land Paradigm” and a new “Predictive Paradigm.”

BLEU score40.0373.27
exact MT match (percent)5.0%34.0%


Have We Been Using MT Wrong?