[developers] Punctuation and "-default-les" type mapping in PET/ERG

Dan Flickinger danf at csli.stanford.edu
Fri Mar 21 12:30:00 CET 2008

Hi Richard -

> I'm trying to use PET/ERG with the "-default-les" option
> to convert SMAFs with tokenizations and POS tags to RMRSs
> in a mode that's robust against missing lexical entries.
> Now I have the following problem: When a token includes
> punctuation, the type mapping doesn't seem to be happening.

I just recently understood the cause of this problem, which is due
to an error in the ERG's PET settings.  I'll check in a corrected
version of the grammar soon, but in the meantime, you can get rid 
of the problem with the following change:

  Delete the setting for "generic-le-suffixes" in erg/pet/common.set.  

This was a pre-punctuation era attempt to constrain proposed generic 
entries to only those with regular inflectional morphology, and is 
now both unnecessary and wrong, given how the ERG treats punctuation.  

Sorry about the confusion this has caused several of you.

> I also have another, possibly unrelated, issue. Which is that,
> even for the sentences that do work, I get a lot of odd warning
> messages like this:
> ; WARNING: failed to unify new path-value ( = "No") into fs (type: 
> d_-_no_le)

If these messages persist after you've removed the generic-le-suffixes
setting, then maybe someone else can help diagnose this, or maybe
indeed you can just ignore the warnings.


More information about the developers mailing list