Dear Philipp Koeh, thank you for your help. I computed several factors using an unsupervised algorithm. My intention is to use these factors to generate alternative translations. I know, that the factors help with translation when tested separately. I wanted to find out whether more factors can be combined together in order to increase the performance furthermore. The reordering models that produce zero scores should not cause problems. I rely on EM in MERT to distribute proper weights.
Regards, Miloslav Dne 18.6.2013 17:13, Philipp Koehn napsal(a): > Hi, > > not as an answer to your question, but as a comment to the model you build: > > You define several different reordering models, and the way these are > handled that Moses will score every phrase translation with all of > them. If some of them have no scores, then there will be no reordering > cost, which the model generally prefers. If you want to have a > different behavior, then you would have to implement it. > > -phi > > On Mon, Jun 17, 2013 at 9:32 PM, Ing. Miloslav KONOPIK Ph.D. > <[email protected]> wrote: >> Hello, >> >> I would like to kindly ask you for your help with the following problem. I >> am using Moses (latest version, checked out today 6/17/2013) compiled with >> –max-factors=16. I use 8 factors and I still get the following exception: >> >> Exception: bitset::set. >> >> Please, see the complete verbose output: http://pastebin.com/hLZyjVSE >> >> The content of moses.ini is included in the aforementioned listing. >> >> The training is done like this: >> >> train-model.perl \ >> --external-bin-dir /software/moses-1.0/bin \ >> --cores 16 \ >> --corpus $SCRATCHDIR/corpus/00train.combined.clean \ >> --root-dir experiment \ >> --f cs --e en \ >> --lm 0:5:$SCRATCHDIR/lm/lm.surf.blm.en:8 \ >> --lm 1:5:$SCRATCHDIR/lm/lm.stem.blm.en:8 \ >> --alignment-factors 1-1 \ >> --reordering msd-bidirectional-fe \ >> --reordering-factors 0-0,1+1-0,1+5-0,1+6-0,1+7-0,1+8-0,1+9-0,1+10-0,1 \ >> --translation-factors 0-0,1+1-0,1+5-0,1+6-0,1+7-0,1+8-0,1+9-0,1+10-0,1 >> \ >> --decoding-steps t0:t1:t2:t4:t5:t6:t7 >> >> Thank you. >> >> Regards, >> >> Miloslav Konopik >> >> _______________________________________________ >> 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
