[developers] Qs: How to check the feature structure under S

Stephan Oepen oe at csli.Stanford.EDU
Tue Feb 13 02:01:19 CET 2007

> not CCing to your students!  This is so long-standing as to be almost
> a feature - my recollection is that originally a representation of
> the new orthography was stored as a feature structure on the relevant
> edge and used when regenerating the trees but this caused some
> problems with the dag recycling which meant it was removed - and not
> replaced by another mechanism.

i actually believe this may be a trivial fix.  my recollection was that
moving to the stepwise application of inflectional rules, the recording
of .orth-tdfs. in the edge now has to be done in a few more places; for
the active parser, at least, i think i have that fixed now.  thus i see
no need for a change in the display (or definition of what effect these
rules should have on parsing results).

i too remember subtle issues with unification contexts and the like.  i 
suspect the current solution (recording the resulting orthography as an
AVM rather than a simple string) has to do with process-unifications(),
which is called by make-orth-tdfs(), being unhappy within a context.  i
actually had to change that last year, and as i find the time to merge
back pending LOGON changes, i will email more about this.  in summary,
i think we ought to be able to work this out.

finally, dag recycling could also be an issue, if we were to hold on to
edges with dags from the recycling pool, reset the pool, and then reuse
any of the dags associated to the edge.  but that would affect all AVMs
in this scenario, and i believe we (aim to) prevent this situation by
disabling dag recycling in interactive mode.

                                                          best  -  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 csli.stanford.edu; oe at ifi.uio.no; stephan at oepen.net ---

More information about the developers mailing list