[logon] HandOn release schedule
Stephan Oepen
oe at ifi.uio.no
Tue Nov 18 15:14:57 CET 2008
> I wouldn't say no to another round.
yes, i suspected that much. which developer wouldn't :-)? the gate is
open again (i think, let me know if SVN access control does not work as
i think it should). i will finally lock the HandOn release on 24-nov,
and then ask everyone to retrieve the final snapshot and re-test.
> BTW, one thing we would like to do is change the model weights for jaen.
> Locally, we change them in .tsdbrc. I am not sure which script they
> need to be put in to work: I guess they are referenced by the transfer
> engine itself (so can go in mt.lsp) but also the controller (thus in
> ja2en.lsp)? Can you advise us here?
[...]
> (setf *mt-analysis-weight* 0.2) ; jacy
> (setf *mt-transfer-weight* 0.2) ; mrs.blm
> (setf *mt-realization-weight* 0.1) ; Erik
> (setf *mt-lm-weight* 0.5) ; bnc.blm
> (setf *mt-distortion-weight* 0.0) ; who knows?
> (setf *mt-lfn-weight* 0.0) ; we aren't norwegian-english
> (setf *mt-lnf-weight* 0.0) ; or visa-versa
these are the (naive) re-ranker parameters, i.e. get used at the point
when fan-out (for one input) is complete. `ja2en.lisp' is the proper
place to set these, i.e. the JaEn [incr tsdb()] controller process. i
have checked the values above into SVN just now.
i believe the LFN weights are quite effective for NoEN, so you might
consider training GIZA++ word-to-word models on your language pair at
some point. i suspect you have all the pieces readily available ...
cheers - oe
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+++ Universitetet i Oslo (IFI); Boks 1080 Blindern; 0316 Oslo; (+47) 2284 0125
+++ CSLI Stanford; Ventura Hall; Stanford, CA 94305; (+1 650) 723 0515
+++ --- oe at ifi.uio.no; oe at csli.stanford.edu; stephan at oepen.net ---
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
More information about the logon
mailing list