normalizeTable > > I use a more straightforward method (below) which yields in different > output, could someone elaborate? > > > > Mycode: > > Probability(f|e) = Count(e,f) / count (e) >
The code that normalizes the counts to probabilities is in TTable::normalizeTable. But, for the initialization, it uses a uniform distribution that isn't computed with this code (giza only stores the counts for pairs of words that actually cooccur in the training data, to save space, since doing otherwise would require keeping an |F_E|x|V_F| matrix in memory. Thus if you want a true uniform distribution, where any word can translate into any word, it must be set up artificially rather than using this code). But, in the end, model 1 is special and doesn't really care what you initialize it too- EM will always converge to the same solution for this model. --Chris _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
