Hi Folks

I've updated the phrase-based Moses toolchain so that word alignments 
are included by default. The hierarchical tool chain will be updated 
shortly in the same fashion.

If you're using EMS and don't want word alignments, then add the 
following to the TRAINING section:
include-word-alignment-in-rules = no

For train-model.perl, you switch off word alignments with
--no-word-alignment

For scoring, you switch them off with
--NoWordAlignment

And for phrase-based binarisation with
-no-alignment-info.

(Some day, the arguments might be consistent, but for now best to let 
EMS worry about it ;-))

As ever, if you notice anything broken, please let moses-support know,

cheers - Barry

On 14/11/12 19:56, Hieu Hoang wrote:
> The decoder flags for word alignments have been cleaned up a little. 
> They were overlapping or didn't work.
>
> These are the ones that work:
>    -print-alignment-info-in-n-best [true/false]
>    -alignment-output-file [filename]
>
> The following were deleted:
>   -use-alignment-info
>      Can be inferred if either of the above switches are used
>   -print-alignment-info
>      Didn't do anything
>
> The switch
>    -include-alignment-in-n-best [true/false]
> was renamed
>    -include-*segmentation*-in-n-best [true/false]
> to better reflect what it actually does
>
> Sorry if it breaks anyone's code
>
>
> _______________________________________________
> 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