Statistical Machine Translation

Statistical machine translation (SMT) was proposed in the 1990s. This translation method uses the source and target sentence pairs and has a translation model and a language model. A decoder uses these models to output a target sentence with the maximum probability. The following is an example of English-Japanese SMT [12].

$$J = argmax_{j}P(j\vert e)$$ (1)

$$\simeq argmax_{e}P(e\vert j)P(j)$$ (2)

Here, $P(e\vert j)$ means the English-Japanese translation model, and $P(j)$ means the Japanese language model. The translation model has probabilities of Japanese words translated into English words. These probabilities are calculated from the English and Japanese sentence pairs. On the other hand, the language model has probabilities of Japanese word strings. The decoder selects the Japanese sentence by referring to the translation model and language model. Statistical machine translation was initially word-based. Recently, though, it has become phrase-based because of the translation performance.