[developers] generation bug in Agree - "are not permitted to look you."

Spencer Rarrick spencer.rarrick at gmail.com
Tue Dec 3 05:44:35 CET 2013

Stephan, thanks for these examples that show why strategies 2 and 3 in my
email may have undesired consequences when dealing with underspecification
in generation.

Perhaps, in some cases it may be desirable to explicitly forbit use of an
optional argument?  I suppose in this case skolemizing to a string that
does not appear elsewhere in the input MRS would accomplish this, so long
as no unskolemized variables elsewhere. It seems like this would be tricky
to enforce, however, when combined with desirable underspecification such
as in the dog, cat, chase example. More concretely, as chase_v_1 has an
optional ARG3, It would be nice to be able to forbid in the input MRS
unification of that ARG3 with arguments in other EPs, but still have
underspecified ARG1 and ARG2 (in this particular case, grammar rules would
likely prevent this from happening anyways, but I'm sure an example exists
where this wouldn't be the case).


On Sat, Nov 30, 2013 at 6:38 AM, Stephan Oepen <oe at ifi.uio.no> wrote:

> hi spencer, glenn, and all,
> > 2. During PRED lookup we could require that the number and names of
> > arguments in candidate LEs matches exactly with those in the input MRS
> EPs.
> i believe this strategy would be incompatible with optional
> semantic arguments.  for example, the ERG treats quite
> a few verbs as optionally transitive, e.g.
>   (1) He exaggerated.
>   (2) He exaggerated the problem.
> both use the same predicate _exaggerate_v_1.  i would
> think one should be able to generate (1) from either of
> the following inputs
>   _exaggerate_v_1[ARG0 e, ARG1 x]
>   _exaggerate_v_1[ARG0 e, ARG1 x, ARG2 u]
> > 3. We could generate and assign skolems for argument positions not
> > constrained by the input MRS (e.g. ARG2 on "look").
> this strategy would mean that argument positions cannot
> be underspecified in generator inputs.  i think one might
> well want to be able to generate from something like:
>   _dog_n_1[ARG0 x1]
>   _cat_n_1[ARG0 x2]
>   _chase_v_1[ARG0 e]
> > Anyway, thanks to anyone who has made it through this incredibly
> long-winded
> > email. If you have ideas about what the correct way to fix this bug,
> > suggestions would be greatly appreciated.
> i believe there is no running away from a post-generation
> semantic compatibility test, e.g. MRS subsumption, so as
> to deal with various types of underspecification in inputs
> to the generator.
> best wishes, oe
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> +++ Universitetet i Oslo (IFI); Boks 1080 Blindern; 0316 Oslo; (+47) 2284
> 0125
> +++    --- oe at ifi.uio.no; stephan at oepen.net; http://www.emmtee.net/oe/ ---
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20131202/66b30fe3/attachment.html>

More information about the developers mailing list