[developers] [pet] unpacking
yzhang at coli.uni-sb.de
Fri Jul 13 09:05:51 CEST 2007
Thanks Dan for figuring out the cause of the problem. And I am looking
forward to the new release of the grammar :-)
On 7/12/07, Dan Flickinger <danf at csli.stanford.edu> wrote:
> Hi Sergio and Yi -
> I tracked down a bug in the ERG which led to this crashing of PET for the
> example you gave - in fact, also for the simpler example
> "We explained to kim why abrams arrives."
> The source of the problem is that "why" is not properly constraining its
> specifier (as in "exactly why") properly. In particular (ignore the rest
> of this paragraph if you don't care about the grammar-internal behavior),
> the phrase "to kim" is incorrectly being allowed as a specifier, and this
> unexpected demand on "to kim" as a specifier can lead to nontermination in
> the parser, due to the analysis of non-local dependencies as in "too
> to ignore" where "too" licenses a VP/NP somewhere later in the sentence,
> which means it has to pass up this demand for a complement to the phrase
> that "too" is a specifier for. It is this general mechanism that passes
> deferred complements up from specifier to phrase that is not yet
> debugged, and sadly this can in principle still lead to nontermination for
> certain ordinary-looking sentences.
> I have corrected this flaw in the new version of the ERG which I am
> to release tomorrow, and I'm grateful that you helped to identify such an
> embarrassing black hole in the grammar before I issued the release.
> Sorry to have contributed to the bewildering behavior. Please try the new
> version of the grammar once I check it in.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the developers