Hi, there are no guarantees that more data will lead to better performance. It typically does, but the type of data also matters, and there is a significant amount of variance so that any single alignment decision may be best handled under any data condition / setting.
-phi On Thu, Aug 25, 2016 at 9:37 PM, Fathima Farhath Farook < [email protected]> wrote: > Dear All, > currently i am testing ( using EMS) the moses on language pair of sinhala > - tamil. where i keep on adding data and doing testing. ( still the number > of sentence pair is in 2000 - 3000 range ) > > when comparing two iterations, one noticeable point was that, with extra > added data the the alignment what was there in previous run become > unaligned in the latter. which resulted to get a different translation - > which is not correct. ( same sentence pair which was shown with word > alignment in previous run was categorized as mismatched in the second run - > in concordance view ) . can anyone explain me the reason behind this > behavior and how to overcome this. > > Thank you > > -- > Regards, > > Farhath Farook > > > <http://farhathforever.blogspot.com/> > > _______________________________________________ > 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
