Hi everyone, I have a doubt about how rule scoring for GHKM string-to-tree rules is implemented in the current version of Moses. Are the scores normalized over the source side of the rules only (so rules 1) and 2) would be competing) or source side and aligned target non-terminals (so 1) and 2) are not competing)?
1) X/NP -> der X Hund, the ADJ dog 2) X/NP -> der X Hund, the NN dog What is the reasoning behind the implemented normalization? Thanks a lot for the help! Cheers, Fabienne _______________________________________________ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support