[developers] Dominance Constraints

[Stephan Oepen]

dear all,

years ago, we had a BSCW workspace for PET, and at some point marco
had posted a relevant proposal (which, sadly, was not followed up much
in the forum discussion at least).  i am about to shut down the workspace,
hence forward the message from marco below.

all best, oe


======= 7-jul-04 posting by marco =======


As two previous papers have shown (links below), a large fragment of MRS can
be translated into dominance constraints, an alternative formalism for the
representation of underspecified semantics.

One of the benefits of this translation is that the highly-efficient solvers
available for dominance constraints can be used to solve MRS descriptions.
The fastest solvers available for dominance constraints are several orders
of magnitude faster than the MRS solver shipped with LKB.

Secondly, as we would claim, the translation from MRS into dominance
constraints also constitutes a correctness test for MRS descriptions: In the
Redwoods treebank, seemingly all MRS descriptions that could not be
translated into dominance constraints were incomplete.

We (Alexander Koller, Stefan Thater, and myself, all at Saarbrücken) would
be interested to contribute an add-on to PET by which it would be able to
produce not only MRS, but also dominance constraints as the output of its
semantics construction. This could be done either as a standalone
application (that would be used as a post-processing step for PET in a
pipeline architecture), or integrated into PET itself, as an alternative
output format. As the result, we envision a software package containing both
PET and one of our dominance constraint solvers, which would combine the
world's fastest HPSG parser with the world's fastest solver for
underspecified semantics. :-)

I am writing this note mostly because I am interested in comments on this
plan, and to hear whether anyone of you would be interested in such a
"product". It would also be interesting to hear about any technical
obstacles that we might not have thought about.

    Thanks,
    Marco

PS: Here are the papers about the translation from MRS into dominance
constraints:

http://www.ps.uni-sb.de/Papers/abstracts/mrs_dom.html
http://www.ps.uni-sb.de/Papers/abstracts/mrs_eval.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20060224/177d552e/attachment.html>