hi dong
combining phrase-based, hierarchical and syntax models is not currently
possible in Moses. If you want to do this, there is significant
programming to be done.
I think other people have done this but as a system combination problem
where use the n-best lists or the output lattice to merge the result
from system.
On 17/03/2012 19:16, dongxinghua0213 wrote:
hi ,
all ,
in moses toolkit ,thare is hierarchical phrase-based model
,string-to-tree model and string-to-tree model ,
we want to combine them together. In other words, we can learn three
different rule tables from parallel texts using the three models.
when decoding ,we can obtain partial hypothesis , which covers input
sentence span [i,j],using the rules from different rule tables, so
we want to use these rules together for every span [i,j] and the
decoding process could be derived mixedly, but the rules from
different tables should be weighted differently . The mert process
does't work in this case . how we can obtain the Optimized weights ?
thank you very much ! best wishes !
_______________________________________________
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
