[developers] More ACE/LKB divergences (underspecified generation)

[Woodley Packard]

Hi again,

> If we restrict the gender to m-or-f [PERS: 3 NUM: sg GEN: m-or-f]
> ACE generates only things with exactly the same semantics:
> They sleep
> (i actually generates the others then rejects them with the heart-warming message: 
> NICE TRY! `She sleeps.' has incompatible semantics:
> NICE TRY! `He sleeps.' has incompatible semantics:
> 
> It seems to me that currently ACE is checking for equality, not subsumption.

Hmm, you have led me to uncover a bug :-/.  Here's what's going on.  As per the St. Wendel discussion you cited, the status of that subsumption check is slightly murky, because there are a couple of ways it could be done, specifically: using the grammar-internal type hierarchy or using the SEMI hierarchy.  The objects to be compared are MRSes, so the elements in question ought to be governed by the SEMI, but traditionally the comparison has actually been made by a round-about procedure involving VPM'ing into the grammar internal type hierarchy and doing the subsumption tests there.  At some point I made the decision to do the comparison without VPM'ing in ACE, but it seems I didn't get all the way there, so at the moment the system seems to be comparing external elements using the grammar internal hierarchy (a bad idea and perhaps formally undefined, but apparently mostly functional).  The type m-or-f does not exist in the ERG's internal hierarchy, so it is treated as subsuming only itself.

The plan is now to fix the post generation subsumption test to use the SEMI hierarchy, which includes the m-or-f type, but I am a bit worried that fixing that will cause the ship to spring a leak somewhere else... we'll see :-)

-Woodley
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20150304/dca8fbaf/attachment-0001.html>