Towards Knowledge-Based Neural Machine Translation

Speaker:
Deyi Xiong
Abstract:
Data-driven machine translation has shifted from traditional statistical machine translation to neural machine translation (NMT) powered by deep learning. Although translations generated by NMT systems are now much better than those by SMT in some languages, it is still difficult for NMT to handle discourse-level machine translation, translation of sentences with complex linguistic patterns and so on. Prior and external knowledge is widely considered helpful on these issues. However, incorporating external symbolic knowledge into NMT is non-trivial. In this talk, I will discuss the connections among data, knowledge and NMT models, as well as our recent efforts in exploring external knowledge in NMT.
Length:
00:55:09
Date:
05/11/2018
views:

Images:
Attachments: (video, slides, etc.)
51 MB
109 downloads
608 MB
112 downloads
100 MB
111 downloads
147 MB
106 downloads
321 MB
109 downloads