Using Lexical Reordering Model

[Behrooz Vedadian]

Hi all,

Forgive me if it is a primitive question, yet there is no clear set of instructions to follow in this regard neither in "Jane Users Manual" nor on the web.
The question is -- in essence -- "How do I used the learnt models of lexical reordering with Jane?"

To be able to use lexical reordering, I changed the extraction configuration like this:

extractOpts="--extractMode=phrase-based-PBT \
             --standard.nonAlignHeuristic=true \
             --standard.swHeuristic=true \
             --standard.forcedSwHeuristic=true \
             --standard.maxTargetLength=10 \
             --standard.maxSourceLength=10 \
             --filterInconsistentCategs=true \
             --additionalModels=lrm
             --lrm.maxTargetLength=10 \
             --lrm.maxSourceLength=10 \
             --lrm.bidirectional=true"

Then ran the trainHierarchical.sh script and finally tried to normalize the lexical reordering scores by normalizeLRMScores and finally used local optimization tool as I used it without lexical reordering model (with new phrase score table).

Adding a new cost model name lrm was doomed to fail and as I searched all the cost names (costNames_.push_back) there were nothing related to lexical reordering model. I gathered that the lexical reordering scores are stored along with phrase scores somewhere and It seems to be utterly wrong. It seems that besides the extraction phase which has generated some output related to lexical reordering model, there has been no change in the way Jane does what it does!

Please instruct me as to how use the lexical reordering model.

Thanks,
Behrooz