Hi Souhir I think you have to try the different suggestions to see if there's a difference. I would normally just train a single reordering model on all the data, but I don't know if that is the best approach. The effect of lexical reordering is quite small, so it will probably not matter much which approach you use,
cheers - Barry Quoting Souhir Gahbiche-Braham <[email protected]> on Fri, 15 Feb 2013 15:11:31 +0100: > Hi Barry, > > I'm using moses with two fusionned phrase tables: > phrase1 ||| phrase2 ||| score1.1 score1.2 score1.3 score1.4 score2.1 > score2.2 score2.3 score2.4 2.718 ||| ||| > > My question is : Is it correct to concatenate reordering models ? or > should I let them independ (usage of two reordering models) ? or use > only one? > or should I fusion them as phrase tables? > > Normally if there 's a common phrase between the two phrase-tables, > order should be the same!! that's why I thought to just concatenate > them. > Bests > Souhir > > -- > Souhir GAHBICHE-BRAHAM / PhD Student > > Univ. Paris Sud, LIMSI-CNRS > Bâtiment S, Bureau S12 > 91403 ORSAY (France) > Tél: +33 (0)1 69 85 80 98 > > -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
