Hi all Giza has an option for giving a dictionary in input. According to the doc (README file in giza++ directory) the effect is that when a word can be aligned with its dictionary correspondant, giza does not try to align it with other words. There is now an option in train-model.perl to use it : -dictionary <file>
The file contains the dictionary entries: the foreign word, then the english word, separated by whitespace. So, if your model translates "das ist ein kleines haus" to "this is a small house", you would have a dictionary similar to this: klein small haus house When I tried to use it, I didn't see any significant improvement compared to adding the dictionary entries in the input corpus, but feel free to try too. Prasanth recently asked how to add a dictionary with multi-word entries: this doesn't handle it, only mono-words. Cheers, -- Raphael _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
