Future (Present?) of Machine Translation

It is quite easy to believe that the recently proposed approach to machine translation, called neural machine translation, is simply yet another approach to statistical machine translation. This belief may drive research effort toward (incrementally) improving the existing neural machine translation system to outperform, or perform comparably to, the existing variants of phrase-based systems. In this talk, I aim to convince you otherwise. I argue that neural machine translation is not here to compete against the existing translation systems, but to open new opportunities in the field of machine translation. I will discuss three opportunities; (1) sub-word-level translation, (2) larger-context translation and (3) multilingual translation.

[@ericpa06]

Pretty interesting talk! I think machine translation is probably one of the most important things because… I mean, imagine how much more knowledge people would have if they could read texts written on any language?

[@ericpa06]

The way I see is that.. if we look from just a mathematical point of view... there are toooooo many variables. I mean, you can rearrange 100 words in so many combinations, and even if you look "oh, those words in Chinese usually appears next to this other words, so I'll use this a base to translate the text to English", even this is not enough. Your big data wouldn't be enough, I think. Basically the point is that… The only way I see us having real freaking good machine translation, as good as the best freaking human translation, would be to create a machine that has some sort of real understatement of how the world works. And use this Knowledge in its translations. Basically I think we would have to have strong artificial intelligence. But maybe I'm wrong. I mean, probably when they are building a machine to play Go, someone must have said the same thing "Oh, you need artificial strong I.A", but anyway...

[@aytenbabashova3954]

👏