Study of Machine Translated Segment Pairs

This study compares customized statistical MT (CSMT) to Google’s neural MT (GNMT) for roughly 2,300 segment pairs from 31 professional translators.

The study uses two well-established estimation scores to predict the impact of machine translation on a translator’s productivity. On average, CSMT exactly matched a translator quality 7 times more often than GNMT and reduced the remaining post-editing work by 50%.

Download the free report for the complete results and analysis.