Stastistical Machine Translation of Japanese

Statistical machine translation (SMT) uses large amounts of language training data to statistically build a knowledge base for translating from one language to another. Before introducing this language data, usually in the form of a parallel set of sentences from both languages, the SMT system has no other linguistic information available to it. With supervised SMT, however, additional linguistic knowledge is allowed in addition to the training data. When translating between languages with little or no common linguistic backgrounds, like English and Japanese, using supervised SMT is extremely useful. By giving the system linguistic rules before training on the parallel corpus, the SMT system can build better alignments between words in both languages.