Hi A language model models one language only. You don't make a language model from a parallel text, but from a monolingual text.
Usually you are interested in a language model for the language into which you translate (you don't need to generate any sentence in source language). So, if in your case you want to translate from English to a regional language, you will want a language model in this language, but not in English. The count file is just this: counts, statistical probabilities for each n-gram. There is not much to understand, but anyway you don't need to read it, just feed it to moses. I don't know how to generate a vocabulary file with srilm. If you want a better control on the individual steps of the language model build, you might want to look at this: http://www.speech.cs.cmu.edu/SLM/toolkit_documentation.html Hope this helps Raphael 2011/1/24 nakul sharma <[email protected]>: > hi all, > > i am building a n-gram language model from parallel text of regional > language of india and English. please tell if i need SRILM understand the > regional language or not. it has produced a count file which is not > understandable even after installing the corresponding font. Also i am > unable to generate vocabulary file for this text.Please forgive if question > is too basic > > > > > > -- > Thanks & Regards, > nakul > > > _______________________________________________ > Moses-support mailing list > [email protected] > http://mailman.mit.edu/mailman/listinfo/moses-support > > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
