Macduff Hughes
Thu 17 Nov 2016, 15:30 - 17:00
Informatics Forum (IF-4.31/4.33)

If you have a question about this talk, please contact: Diana Dalla Costa (ddallac)


Deep neural networks for machine translation have produced impressive research results in the last few years (notably from the University of Edinburgh), greatly improving over formerly state-of-the-art statistical methods. Bringing these systems into production for end users presents unique challenges: of scale, speed, features, translation quality, and the ability to diagnose and fix unexpected behavior from deep neural networks. This talk will cover some of those challenges that Google has overcome in converting a large fraction of Google Translate traffic to neural translation, and some of the challenges that still remain.


Macduff Hughes is an Engineering Director at Google, and has led the Translate team there since 2012.