I have done that before; using stems to align and then the full version to extract and score phrases. What I usually do is run the script twice, i.e.
train-factored-moses.perl --last-step 3 --corpus /path/to/stem and then train-factored-moses.perl --first-step 4 --corpus /path/to/full/corpus Or if you use a factored corpus, you can use the alignment-factors flag to specify which factors use to align and then translation-factors flag to tell it which one use to extract and score. I haven't tried the second option but it should work too. -- Carlos A. Henríquez Q. ________________________________ De: Somayeh Bakhshaei <[email protected]> Para: [email protected] Enviado: mar,27 abril, 2010 09:24 Asunto: [Moses-support] changing train-factored-phrase-model.perl Dear All, Hi, I need to change the corpus after running Giza++. I mean i want to use different kind of corpus for these steps: making alignment models and extracting phrases. I should change train-factored-phrase-model.perl. is it enough to rm the old corpus and cp new one like this there? ### MAIN &prepare() if $___FIRST_STEP==1; &run_giza() if $___FIRST_STEP<=2 && $___LAST_STEP>=2; &word_align() if $___FIRST_STEP<=3 && $___LAST_STEP>=3; &get_lexical_factored() if $___FIRST_STEP<=4 && $___LAST_STEP>=4; rm corpus/currcorpus cp .... corpus/newcorpus &extract_phrase_factored() if $___FIRST_STEP<=5 && $___LAST_STEP>=5; &score_phrase_factored() if $___FIRST_STEP<=6 && $___LAST_STEP>=6; &get_reordering_factored() if $___FIRST_STEP<=7 && $___LAST_STEP>=7; &get_generation_factored() if $___FIRST_STEP<=8 && $___LAST_STEP>=8; &create_ini() if $___LAST_STEP==9; ------------------ Best Regards, S.Bakhshaei
_______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
