Hi,

there are a number of different scores in the phrase table.
The phrasal scores are normalized (they add up to one
for each phrase that is conditioned on) - unless smoothing
methods such as GoodTuring and KneserNey are used
(then they add up to less than one).

The lexicalized scores are not normalized.

For interpolation it generally not a good idea to normalize
scores.

-phi

On Mon, Jan 21, 2013 at 12:35 AM, Barry Haddow
<[email protected]> wrote:
> Hi
>
> The phrase probabilities in Moses are normalised. At least they should be.
>
> For interpolation of phrase tables, look in contrib/tmcombine
>
> cheers - Barry
>
>
> On 21/01/13 06:52, M.Osarenejad wrote:
>
> Hi,
>
> I observed that the probabilities in Moses phrase-table are not normalized.
> I want to interpolate two phrase-tables. Is it necessary to normalize the
> scores when interpolating phrase-tables?
>
> Thanks,
>
>
>
>
> _______________________________________________
> 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
>
> The University of Edinburgh is a charitable body, registered in
> Scotland, with registration number SC005336.
>
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to