Hi, I run the same test as the previous email using 1.1 M training data and also in this case I got the same bleu score and the same target sentences when I use the binary and normal SRILM language model
Thanks Marco On Thu, Dec 3, 2009 at 1:31 PM, marco turchi <marco.tur...@gmail.com> wrote: > Hi Philipp and Hieu, > > i run this test: > I build a moses model using 62k training sentences, then I build 2 LMs > using the same data (English side): > 1)the normal srilm > 2)the binarized srilm > > then i translate. > > I got the exactly the same target sentences (I used the "diff" command to > compare them). and Blue score (I did not optimize to speed up the test) > > Now I'm running a new test using 1.1 M sentences... I'll let u know later. > > the small LM value in the previous example I guess it is due to small > amount of training data (as in that case I used only 62k). > > if u have any comment, please let me know :-) > > Thanks a lot > Marco > > > On Thu, Dec 3, 2009 at 7:48 AM, Philipp Koehn <pko...@inf.ed.ac.uk> wrote: > >> Hi, >> >> if you generate an n-best list, the types of the different feature >> functions >> are more transparent, but it is a safe bet that your example "-312.951" >> is the unweighted language model weight. >> >> -phi >> >> On Wed, Dec 2, 2009 at 11:44 PM, marco turchi <marco.tur...@gmail.com> >> wrote: >> > Hi Hieu, >> > I'll check... and I'll let u know. >> > >> > In case this the output of moses using a binarized srilm LM: >> > [total=-40.478] <<-8.000, -74.000, 0.000, -11.209, -4.077, -2.675, >> -12.833, >> > -0.519, -5.449, -312.951, -66.706, -138.406, -42.936, -83.404, 32.997>> >> > which is the LM value? Is it the 10th value? >> > >> > Thanks >> > Marco >> > >> > On Wed, Dec 2, 2009 at 11:23 PM, Hieu Hoang <hieuho...@gmail.com> >> wrote: >> >> >> >> no-one's asked this before but i'll be interested to find out. If you >> >> anything more, let me know >> >> >> >> marco turchi wrote: >> >>> >> >>> Hi, >> >>> just a quick question, can Moses use a binary version of the srilm lm? >> >>> I have built it with the option -write-binary-lm >> >>> >> >>> Thanks >> >>> Marco >> >>> >> ------------------------------------------------------------------------ >> >>> >> >>> _______________________________________________ >> >>> Moses-support mailing list >> >>> Moses-support@mit.edu >> >>> http://mailman.mit.edu/mailman/listinfo/moses-support >> >>> >> > >> > >> > _______________________________________________ >> > Moses-support mailing list >> > Moses-support@mit.edu >> > http://mailman.mit.edu/mailman/listinfo/moses-support >> > >> > >> > >
_______________________________________________ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
