Unfortunately for Model 3/4 you can't get Viterbi probability - Viterbi
decoding for fertility models is not implemented in giza. Viterbi decoding
for fertility model is exponential. So the prob you get in the final
alignment file is an approximation using hill climbing.

If you want to do posterior decoding the best bet is to use HMM model and
implement forward-backward yourself. I don't remember there's an
implementation in giza.

--Q


On Mon, Apr 7, 2014 at 4:14 PM, Nima Pourdamghani <[email protected]> wrote:

> Dear All,
>
> Is there any way to get the posterior probabilities of alignments for
> Model 3/4 in Giza++ / Mgiza++?
> I want to do posterior decoding but I can't get these probabilities.
> Even just the probability of the Viterbi best alignments will help a lot.
>
> Bests
> --
> Nima Pourdamghani
>
> _______________________________________________
> 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

Reply via email to