Being informed about machine translation

There is a lot of hype about machine translation (MT). But we all know that MT, in its current form, generally cannot produce good translations. So how hard is it really for the machine to translate? You might want to head over to Stanford’s free Introduction to Artificial Intelligence class and have a look at lectures 1 (particularly units 9–12), 21, and 22. From what I have seen there, there is little chance that MT will be usable any time soon, although judging from how far (and how fast) AI has come along in other areas, it would be naive to claim that MT will not one day reliably produce good translations — and that day could be much closer than you think. If you have some free time in winter 2012 (January 23 to March 12 or thereabouts), you might want to sign up for Stanford’s free introductory Natural Language Processing class. Or you can try the Natural Language Processing course on SEE (Stanford Engineering Everywhere), which is available right now. In any case, don’t believe the hype but don’t dismiss it outright either; there are ways we can actually learn about it.