[logon] updated logon; strange behavior with ?pet
Berthold Crysmann
crysmann at ifk.uni-bonn.de
Fri Jun 12 08:17:45 CEST 2009
On Thu, 2009-06-11 at 21:16 -0700, Emily M. Bender wrote:
> Thanks, Dan. I do have a semi.vpm file, but I am not currently
> loading it. When I reconstruct the derivations from the old
> profiles, I do indeed see an event as the index. This was probably
> over determined, though, since I don't think I've ever seen this
> problem with the LKB.
>
> It does seem like this is related to Francis's bug report. Perhaps
> that was fixed?
>
I think the bug report was from me. There was a problem with some
versions of cheap (before May 6) being compiled against older vpm code
which has meanwhile been fixed.
See Stephan's mail from May 6.
Berthold
> Emily
>
> On Thu, Jun 11, 2009 at 4:25 PM, Dan Flickinger<danf at stanford.edu> wrote:
> > Hi Emily -
> >
> > There was a possibly related message from Francis on May 21 (subject: pet/lkb inconsistency) where he noted that the top-level index for an MRS was 'u' instead of 'e' under some circumstance. Maybe he can clarify. It does sound like a bug fix, even if an immediately awkward one.
> >
> > In your old profiles, when you reconstruct the derivation for some item, does the feature structure have the verbs' INDEX values typed to 'event'? I would have though so, if you're using the Matrix 'event-relation' for them. If so, they should have always been getting printed as 'e' in the resulting MRSs. In the LOGON universe this external mapping of the MRS is mediated by the variable-property-mapping, typically defined for a given grammar in a file with a .vpm suffix. If you are using such a file in your Wambaya grammar, see what it says about 'event', and maybe compare it with the file 'semi.vpm' in the ERG.
> >
> > Dan
> >
> > ----- Original Message -----
> > From: "Emily M. Bender" <ebender at u.washington.edu>
> > To: logon at emmtee.net
> > Sent: Friday, June 12, 2009 12:12:51 AM GMT +01:00 Amsterdam / Berlin / Bern / Rome / Stockholm / Vienna
> > Subject: [logon] updated logon; strange behavior with ?pet
> >
> > Hi,
> >
> > Today I finally had time to update my logon tree, and have noticed
> > something odd, which I think I have to attribute to this update:
> >
> > All of my pre-update profiles have the ARG0 of verb relations
> > stored as `u', whereas the profiles that I processed today have
> > `e'. I don't think there's anything I've done to the grammar that could
> > have caused this change... and in fact, I've just run a profile
> > with the pre-today version of the grammar and the current
> > logon/pet (both with the old .grm file and with creating a new one;
> > the change was not in flop).
> >
> > This looks like a bug fix to me, but I'm wondering if it was
> > in fact intentional. I don't think I saw any traffic on this list
> > or `developers' that would have indicated it.
> >
> > Emily
> >
> > p.s. I *know* this is a feature, but it's a rather annoying one just
> > now as I'm doing a bunch of work with the Wambaya grammar that
> > requires comparing MRSs. So now I have diffs on *every single
> > item*. The solution will be to run profiles for selected older versions
> > of the grammar with the new software...
> >
More information about the logon
mailing list