The Future of Machine Translation is Unknown, But We’re Still Worried

Translators love to point out these days that machine translation (MT) is at present much less clever than human translation, and only works to an acceptable level in a few applications. Yes, but to the extent that it works, and it does work to some extent in some fields, it works fantastically quickly and cheaply. Therefore, it necessarily replaces human translators to that extent, for purely profit-and-loss considerations. To this extent, human translators become, as many have pointed out, the equivalent of workers who tend automatic looms and the like to take care of the occasional foul-ups of the machinery. The question is, how far will MT go in replacing humans and how fast? We don't know yet.

I would say that the claim that “deep learning” and the like now enables computers to learn how to play chess exaggerates the amount by which MT is replacing and probably will replace humans. That is, it won't happen entirely; for one thing, chess is a game with precise rules that computers can easily follow. It's just necessary to calculate the probably consequences of making a move according to the rules a lot of moves in advance. This seems to be possible now even for Go, which was thought to be too subtle for computers. But that game is also played with precise rules. 

Translation, on the other hand, works with natural languages, which don't follow rules that even "deep learning" can completely learn, I think. The languages we speak are part of human activity and the human brain, which even the most capable computers have not yet exhausted completely. Human life, and language, are not just board games, which I suspect is a technical barrier to computers learning all about them that will not be soon, if ever, conquered.

E© Jon Johanning 2014