Hi Marwa,

The cache-based model, the bitext sampling model, and the cdec
implementation are three different things.  The cache-based model is
described in this paper
<https://www.matecat.com/wp-content/uploads/2013/09/mt-summit-2013-bertoldi-et-al.pdf>
and
there are directions for using it in Moses here
<http://www.statmt.org/moses/?n=Advanced.CacheBased#ntoc1>.
PhraseDictionaryBitextSampling is described in this paper
<https://ufal.mff.cuni.cz/pbml/104/art-germann.pdf> and data can be added
at runtime with an XML tag in the form:
<update name="PhraseDictionaryName"  source="source sentence"
target="target sentence" alignment="0-0 1-1" />

To understand the cdec post-edit support feature and how to get a similar
(better) version in Moses, consider having two phrase tables: a
"background" table that contains initial training data and is static, and a
"post-edit" table that starts empty and has post-edited translations
incrementally added.  In Moses, this can be done with two instances of
PhraseDictionaryBitextSampling (one built on all initial data and one built
on empty files) and a PhraseDictionaryGroup
<http://www.statmt.org/moses/?n=Advanced.Models#ntoc7>.  After each
sentence is post-edited, the new source-reference pair is aligned and added
to the post-edit table.  Each table has a full feature set so each phrase
pair is scored according to both the background and the post-edited data.

cdec just uses a single table that is built on initial data and also
updated with post-edited data.  The post-edit support feature flags which
phrases would still be extracted when considering only the post-edited
data.  Using two phrase tables with full features (the Moses way) has been
shown to perform better.

Best,
Michael


On Wed, Oct 5, 2016 at 9:45 AM, Marwa Refaie <[email protected]> wrote:

> Hi All
>
>
> Does the "age" feature in the CBMT implemented in the Bitextsampling
> (using suffix array) which is equal to the 'Post-edit support"
>  introduced by Denkoweki in cdec ??
>
>
> Any info please !
>
>
>
> *Marwa N. Refaie*
>
>
> _______________________________________________
> 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