It’s hard to distinguish reality from hype. So, we analyzed 4.3 million translated segments to measure how a Context Sensitive approach compares to NMT in terms of TER (the edit distance it took for the translator to confirm the feed) and what we found was… See for yourself and let us know your thoughts!
We tested 17 language pairs with English as the source, translating into Portuguese (Brazil), Italian, Spanish (Spain and Latin America), French (France and Canada), German, Japanese, Dutch, Simplified Chinese, Danish, Russian, Korean, Vietnamese, Thai, and Turkish.
We analyzed metadata from approximately 4.3 million segments between June 22 and December 23, 2023. Of these, about 2.3 million were translated using traditional Neural MTPE, and around 2 million were translated using a Context Sensitive Approach. Microsoft Neural Machine Translation served as the baseline engine for this study.
Download the research and see firsthand how an approach with context sensitivity compares to NMT.
