Statistical machine translation (SMT) is an approach to MT characterized by the use of machine part of the commercial MT market. Progress has been rapid, and the state of the art is a moving target. We refer to Lopez (2008) for a review of the state of the art in SMT.

SMT views translation as a machine learning problem. So it uses a learning algorithm to a large body of previously translated text . The learner is then able to translate previously unseen sentences. With an SMT toolkit and enough parallel text, an MT system for a couple of new language can be built within a very short period of time.  The accuracy of these systems depends on the quantity, quality, and domain of the data.