hi ales

every hypothesis has a reference to manager. And every manager has a 
pointer to
   InputType const& m_source;
the hypothesis has a variable m_currSourceWordsRange which tells you the 
source phrase span.

I would avoid using TargetPhrase::m_sourcePhrase. This was a hack for 
confusion network/lattice decoding and is overdue for a cleanup

On 28/06/2012 18:21, Ales Tamchyna wrote:
> Hi,
>
> I'm implementing a stateless feature function (for the phrase-based
> decoder) and I encountered several problems. Is it possible to get the
> source sentence and source phrase span somehow? The feature (phrase
> sense disambiguation) depends on the context of the source phrase.
>
> If it's not possible, would it make sense to add an additional interface
> for context-dependent stateless functions? I'm also wondering if anyone
> else would benefit from an interface that would allow scoring all
> translation options of a source phrase at once. (We would, but we can
> work around it by caching etc.)
>
> Also, is there a reason why scores by stateless features are computed
> during decoding? If they were computed for each translation option ahead
> of time (once full translation options are created), the decoder could
> use them in the initial pruning. It could also prevent repeated
> computation of feature function values (e.g. our feature is costly to
> compute).
>
> Thanks!
>
> Ales Tamchyna
> _______________________________________________
> 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