Better Document-level Machine Translation with Bayes’ Rule

Thanks to the rapid development of neural sequence-to-sequence models and the availability of large scale datasets, current state-of-the-art machine translation systems have achieved super-human performance for some language pairs such as French <-> English and German <-> English. However, there’s still a big gap between humans and document-level machine translation systems, because humans are much better at capturing the consistency and coherence of the documents. In this talk, I will first review the existing approaches for document-level machine translation and then present our recent work on Better Document-level Machine Translation with Bayes’ rule. If time permits, I will also briefly introduce the research we have been doing at DeepMind related to machine translation and other language processing tasks.