Hi, Sorry for the wrong information.
After I double check the setting used in Pharaoh and Moses, I found that Pharaoh use "grow-diag-final" to obtain the bidirection word alignment while Moses use "grow-diag-final-and". This cause the phrase tables extracted by two systems obviously different. So, It's the reason. Another new question for me is that: Do the Pharaoh and Moses use the same phrase-extract binary? Or phrase-extract is the updated version in Moses suit? Thanks guys, -Hongfei 2007/11/13, Philipp Koehn <[EMAIL PROTECTED]>: > > Hi, > > the features that typically get better performance > with Moses are the lexicalized reordering and the > support for higher order language models (I think). > With the same features, performance should be > very similar. > > -phi > > On 11/12/07, Hieu Hoang <[EMAIL PROTECTED]> wrote: > > 5 BLEU points !? i wish. > > > > there are small differences between moses & pharaoh which usually makes > > moses slightly better (less than 0.5 BLEU) for traditional phrase based > > decoder. i recent paper @ IWSLT compared the difference between the 2 > > systems > > > > http://www.cs.ust.hk/~dekai/library/WU_Dekai/ShenLoCarpuatWu_IWSLT2007.p > > df > > u can get better results by adding linguistic information, which is the > > reason for moses > > > > > > Hieu Hoang > > www.hoang.co.uk/hieu > > -----Original Message----- > > From: [EMAIL PROTECTED] > > [mailto:[EMAIL PROTECTED] On Behalf Of corporalist > > corporalist > > Sent: 12 November 2007 08:58 > > To: [email protected] > > Subject: [Moses-support] About the comparison between Pharaoh and Moses. > > > > > > Hi, guys, > > > > > > Anyone have ever done some comparison between Pharaoh and Moses > > in open data set? > > > > I wonder how much difference is reasonable. > > > > I have tried a comparison experiment on NIST05 Chinese-English > > test using FBIS. > > > > My results showed that Moses can outperform Pharaoh by 5 Bleu > > score. > > > > Is this difference reasonable? > > > > Anyone did similar comaparison pls kindly tell me your results. > > > > -fei > > > > _______________________________________________ > > 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
