Translation models and language models are completely independent. A translation model seeks to answer the question 'How likely is a particular translation?'. A language model seeks to answer the question 'How likely is a phrase/sentence?' i.e. if your language model is English - 'How good English is the output?'.
You only ever need once train a language model for any particular language. You should be able to reuse it with any language pair. Obviously, the more data you train it with the more representative that language model will be. If you are seeking to do domain specific translation it would obviously make sense to train your language model from the same domain. Quoting João Graça <[EMAIL PROTECTED]>: > Hello, > > I have a question, is the language model used when training a factored > model, besides being place on moses.ini. It seems that all the values > available on the translation table don't depend on the language model. > My question is in case we want to try the same model with different language > models we just have to train a factored model once. Then we can manually > edit the moses.ini and change the language model by hand to make the tests > (and run tunning). > > I am right? > > Thanks a lot, > > Joao Graca > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
