[developers] active/passive parser bug

Dan Flickinger danf at csli.stanford.edu
Tue May 3 12:02:35 CEST 2005

Hi Ben and Stephan -

Thanks for the nice detective work.  The mismatch in the behavior of the
two parsers was due to a slightly stupid error in the lexical entries for
a small set of verb-particle guys that require the particle to always follow
the noun.  Because I chose an unfortunate name for that lesical type
(v_np_particle_only_le), I misled myself into believing that the particle
should be constrained via the OCOMPKEY attribute (that is, the one for
the second complement), but in fact particles are (almost) uniformly
constrained instead via the COMPKEY attribute.  So the lexical entry
see_around_v1 (for "I'll see you around" but "*I'll see around you" on
the intended reading) restricted the OCOMPKEY to be _around_p_sel_rel,
but that OCOMPKEY attribute was not reentrant with anything in the actual
feature structure.  Thus the active parser, using the chart dependency
constraints, serenely excluded see_around_v1 when parsing
  Abrams sees Browne as being a competent manager.
but the passive parser, apparently not applying those chart dependencies,
introduced this item into the chart, and given the error in the lexical
entry see_around_v1, happily built 'see Browne as' using it.

I'll check in the corrected lexical entries for the offending eight entries
later today.


> hi ben,
> > Items #239 and #241 of the CSLI test suite produce many more parses when 
> > processed in passive rather than active mode. This would appear to be a 
> > bug.
> it is in principle possible for the two parsers to differ in the number
> of results found, albeit not desirable.  the active parser includes the
> `chart dependency' filter originally proposed by dr. mueller (for all i
> know, that is).  checking `chart dependencies' prior to parsing aims to
> suppress search doomed to fail, but of course there is room here for a
> grammar to go astray and accidentally suppress intended analyses.  one
> way of testing this hypothesis would be forcing 
>   (defparameter *chart-dependencies*
> to nil after loading the ERG.  --- i just tried PET and get:
>   (1) `Abrams sees Browne as being a competent manager .' [0] --- 2 [...]
> so at least ulrich and i are on the same side here :-).
>                                                          best  -  oe
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> +++ Universitetet i Oslo (ILN); Boks 1102 Blindern; 0317 Oslo; (+47) 2285 798
> +++     CSLI Stanford; Ventura Hall; Stanford, CA 94305; (+1 650) 723 0515
> +++       --- oe at csli.stanford.edu; oe at hf.uio.no; stephan at oepen.net ---
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

More information about the developers mailing list