This is rather interesting, apparently they have software now that approaches translating by learning from the efforts of other translators. "Software learns to translate by reading up" talks about this development. This sort of learn from experience approach works exceptionally well at recognizing spam, which I realize is not quite related, but it is the same sort of language recognition problem that the traditional hard coded rule sets have been failing at. It will be interesting to see how well this sort of thing works.
