Phrase-based vs. Neural MT: Which Helps Translators the Most?

Track: TAUS | TS4 |
Thursday, November 2, 2017, 4:30pm – 5:15pm
Held in: Mission City Ballroom 4
Alessandro Cattelan - MateCat
Host: Anne-Maj van der Meer

According to the infamous announcement from Google, neural machine translation (NMT) matches the quality of human translation. We quickly realized that it didn’t live up to the hype. However, how helpful is NMT for professional translators? We’ve been collecting statistical data on thousands of translations carried out using the phrase-based machine translation (PBMT) versus NMT. We measured post-editing effort and time to edit for a dozen language pairs, analyzing the most common issues with both systems. We’ll describe how the data we collected and the results of our research can help you make informed decisions on which technology to adopt.

Takeaways: Attendees will get an understanding of the actual benefits of using PBMT or NMT and how to measure the impact of MT on production.