Hi, "parse relaxation" in the Moses pipeline means adding more internal nodes by tree binarization or more enriched representations inspired by Zollmann/Venugopal's syntax-augmented model.
It does not really matter when the parse relaxation is carried out. In the experiment.perl pipeline it is done just before the rule extraction. -phi On Wed, Mar 13, 2013 at 8:23 AM, Ken Fasano <[email protected]> wrote: > > > Thank you, > > Ken Fasano > > Begin forwarded message: > > From: "Ken Fasano" <[email protected]> > Date: March 13, 2013, 11:21:41 AM EDT > To: <[email protected]> > Subject: Relaxed parsing, and tree-based model training parameters > > I’ve got a couple of questions regarding the tree-based model: > > 1. Relaxing Parses > I find C++ code for this in mosesdecoder/phrase-extract. When is relaxing > run? After parsing and before training? How is it run? > > 2. (String or tree)-to-(tree or string) > > Does the –corpus parameter in train-model.perl point to the annotated file, > i.e. from BitPar? If so, is the –xml parameter required? If there are two > annotation files, i.e. tree-to-tree, is it enough to use the parameters > --source-syntax and/or --target-syntax? With tree-to-tree, is there only one > parameter to indicate both annotated files, so that they both have to have > the same base name? > > > > Thank you! > > > _______________________________________________ > 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
