Statistical machine translation

Translation treated as a probability problem: a system learns from large collections of translated sentence pairs which output is most likely for an input, instead of following hand-written rules. Phrase-based systems are the best-known kind.