Hi,

In the msd-fe model, there are three possible prodictions:
- p(monotone|f,e)
- p(swap|f,e)
- p(discontinuous|f,e)
Each of the predictions has a separate weight (which is a bit weird, but hey).

There are six weights, because there are two models: one backward
looking (how does this phrase pair reorder with respect to previous
phrase), and one forward looking (how does next phrase pair reorder
with respect to this one).

-phi

On Mon, Dec 8, 2008 at 11:20 AM, Yee Seng Chan <[EMAIL PROTECTED]> wrote:
> Hmm.. I correct myself. After flipping though step 7 of the training perl
> script, I think the first 3 weights are probabilities reflecting how the
> previous target-phrase (extracted) is aligned to the source sentence, while 
> the
> next 3 weights are probabilities for the next target-phrase.
>
> In particular, they are: previous_mono, previous_swap, previous_other,
> following_mono, following_swap, following_other.
>
> Is my understanding correct?
>
> Cheers,
> Yee Seng.
>
>
> _______________________________________________
> 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