Keynote Kyunghyun Cho
Presentation
Abstract
Neural machine translation is the third paradigm of machine translation technology after the rule-based paradigm and phrase-based statistical one. As machine translation evolves from one paradigm to another, we have observed the declining use of linguistics either surprisingly or unsurprisingly depending on whom you talk to. In this talk, I will briefly review the basics of neural machine translation, and continue on to discuss some extreme examples of such trend in declining use of linguistics in machine translation, including fully character-level translation and multilingual translation, both of which were not considered possible or sensibly until very recently. I conclude the talk by describing my perspective on how linguistics or our prior knowledge of natural languages and machine translation may and should be incorporated into machine translation.