You can use the --giza-option to tell GIZA to use a different model, but I don't believe the moses scripts can be told look for the nonstandard file that GIZA will generate in this case (they are named differently for the kind of model used). But, this wouldn't be hard to make this configurable or to hard code it. Also, when you tell GIZA to use model 1, you have to tell it to print alignments (model1dumpfrequency is the name of the parameter i believe), which is simply the iteration number when you want the alignments produced.
Also, the HMM alignment model is very similar to the IBM models- it just models dependencies between source word positions (whereas model 4, for example, models dependencies between target word positions). HMM alignments are really quite good, and much less expensive to produce than the higher IBM models. Chris On Fri, May 16, 2008 at 11:39 AM, Sanne Korzec <[EMAIL PROTECTED]> wrote: > > > > > Hi mailing, > > Is there a way to force GIZA to only train IBM Model I during training (or > 2 or 3 for that matter)? I saw that the option: --giza-option in > train-phrase-model.perl might be what I needed, but the example in the > tutorial only talks about number of iterations. Does this mean that the > option m1=9,m2=0,mh=0,m3=0,m4=0 will only train 9 iterations on model I and > then quit? > One final thing, why then are HMMs (mh) in this list, I thought HMMs use a > different model which is not to be combined with IBM models. > > Regards, > Sanne > > > > > > _______________________________________________ > Moses-support mailing list > [email protected] > http://mailman.mit.edu/mailman/listinfo/moses-support > > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
