Hello, Thank you Felipe (and Christian and David who replied to me personally). As Christian pointed out, there is (or at least was at one point) a difference between the 2003 paper and the actual implementation (see http://www.mt-archive.info/MTMarathon-2010-Hardmeier.pdf).
/Markus 2011/3/12 Felipe Sánchez Martínez <[email protected]> > Hi, > > It is explained in this paper, sec 4.4: > http://www.isi.edu/~marcu/papers/phrases-hlt2003.pdf > > Cheers > -- > Felipe > > El 11/03/11 05:58, Markus Saers escribió: > > Hello, >> >> I'm wondering about how the lexical scores in the phrase table are >> calculated. I have some memory of hearing someone say something like >> "It's just IBM model 1 scores for the phrase pair", which could be >> interpreted in so many ways... and now that I look at the Moses >> documentation online it seems to be calculated based on the lexical >> translation table, which in turn seems to be extracted from the >> alignments. If so: does this take the alignment scores into account, or >> is every alignment point counted as one occurrence of the word pair? So, >> how is it actually calculated (if it's described in a paper, a reference >> would be greatly appreciated as well)? >> >> Kindly >> Markus >> >
_______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
