it's also used the cache the LM scores and other scores (word penalty,
phrase-penalty) so that they don't have to be recalculated every time the
translation rule is used.

I think, for this reason, the phrase table should always be the last
feature function to be loaded

On 5 February 2013 09:44, Kenneth Heafield <[email protected]> wrote:

> Hi,
>
>         All target-side phrase table entires are scored by the language
> model
> during loading.  These scores are used to prioritize phrases and to exit
> language model adjustment early when it determines that the existing
> scores of words in phrases will not change.
>
> Kenneth
>
> On 02/05/13 09:38, bougares wrote:
> > Hi all,
> >
> >
> > In moses code and before loading Phrase tables there is a test in
> > "StaticData::LoadPhraseTables()" if  language models are already loaded
> > (it's also written that LM must be loaded prior to loading phrase
> tables).
> >
> >
> > Is there any explanation for this order of loading ?
> >
> >
> > Thanks.
> > Fethi.
> > _______________________________________________
> > 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
>



-- 
Hieu Hoang
Research Associate
University of Edinburgh
http://www.hoang.co.uk
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to