Hi Anoop, I am not aware of how to incorporate quality score of parallel sentence during SMT training. But, you may think to create 2 parallel corpus (PC): PC1: Select all source sentence with their best translated sentence (let's say in-domain) PC2: Rest of the translation with their source sentences (let's say out-domain)
Then you can combine translation models (obtained by training SMT over the 2 parallel corpus) using different approaches ( http://www.statmt.org/moses/?n=Advanced.Domain#ntoc1) Another way to incorporate quality score can be to create dense features. You may have a look at our recent paper ( http://www.statmt.org/wmt15/pdf/WMT25.pdf) on APE at WMT 2015. We have discussed different feature, if it is of any help to you. On Fri, Sep 18, 2015 at 8:35 PM, Anoop (അനൂപ്) <[email protected] > wrote: > Hi, > > Suppose in a parallel corpus, multiple translations of varying quality are > available for every sentence along with a quality score. Is there any way > to incorporate these quality scores during Moses training? Pointers to any > papers which have discussed such ideas of weighting sentences in a parallel > corpus would be useful. > > Regards, > Anoop. > > -- > I claim to be a simple individual liable to err like any other fellow > mortal. I own, however, that I have humility enough to confess my errors > and to retrace my steps. > > http://flightsofthought.blogspot.com > > _______________________________________________ > Moses-support mailing list > [email protected] > http://mailman.mit.edu/mailman/listinfo/moses-support > > -- -Regards, Rajen Chatterjee.
_______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
