Hi, The phrase-table file has a score vector of 5 numbers. Two of them are lexical weights.
If I understand it correctly, the lexical weights of a phrase pair are computed by use of the word alignments within the phrase pair. But in a large corpus of sentences we might encounter the same phrase with a different word alignment, thus changing the value of the lexical weight. How are these combined into the final translation table? Is this simply a matter of averaging them or is there some more sophisticated method used? Regards, Sanne _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
