19.325 conference on Natural Language and Knowledge Representation

From: Humanist Discussion Group (by way of Willard McCarty willard.mccarty_at_kcl.ac.uk>
Date: Sat, 8 Oct 2005 08:22:41 +0100

               Humanist Discussion Group, Vol. 19, No. 325.
       Centre for Computing in the Humanities, King's College London
                     Submit to: humanist_at_princeton.edu

         Date: Sat, 08 Oct 2005 08:05:09 +0100
         From: Jana Sukkarieh <jana.sukkarieh_at_clg.ox.ac.uk>
         Subject: Natural Language and Knowledge Representation ( 2nd CFP)

Special Track at FLAIRS 2006


Holiday Inn Melbourne Oceanfront, Melbourne Beach, FLORIDA, USA

MAIN CONFERENCE: 11-12-13 MAY 2006

Special track web page: http://users.ox.ac.uk/~lady0641/Flairs06_NL_KR
Main conference web page: http://www.indiana.edu/~flairs06


We believe the Natural Language Processing (NLP) and the Knowledge
Representation (KR) communities have common goals. They are both concerned
with representing knowledge and with reasoning, since the best test for the
semantic capability of an NLP system is performing reasoning tasks. Having
these two essential common grounds, the two communities ought to have been
collaborating, to provide a well-suited representation language that covers
these grounds. However, the two communities also have difficult-to-meet
concerns. Mainly, the semantic representation (SR) should be expressive
enough and should take the information in context into account, while the KR
should be equipped with a fast reasoning process.

The main objection against an SR or a KR is that they need experts
to be understood. Non-experts communicate (usually) via a natural language
(NL), and more or less they understand each other while performing a lot of
reasoning. An essential practical value of representations is their attempt
to be transparent. This will particularly be useful when/if the system
provides a justification for a user or a knowledge engineer on its line of
reasoning using the underlying KR (i.e. without generating back to NL).

We all seem to believe that, compared to Natural Language, the existing
Knowledge Representation and reasoning systems are poor. Nevertheless, for a
long time, the KR community dismissed the idea that NL can be a KR. That's
because NL can be very ambiguous and there are syntactic and semantic
processing complexities associated with it. However, researchers in both
communities have started looking at this issue again. Possibly, it has to do
with the NLP community making some progress in terms of processing and
handling ambiguity, the KR community realising that a lot of knowledge is
already 'coded' in NL and that one should reconsider the way they handle
expressivity and ambiguity.

This track is an attempt to provide a forum for discussion on this
front and to bridge a gap between NLP and KR. A KR in this track has a
well-defined syntax, semantics and a proof theory. It should be clear what
authors mean by NL-like, based on NL or benefiting from NL (if they are
using one). It does not have to be a novel representation.


   For this track, we will invite submissions including, but not limited to:

    a. A novel NL-like KR or building on an existing one
    b. Reasoning systems that benefit from properties of NL to reason with NL
    c. Semantic representation used as a KR : compromise between expressivity
and efficiency?
    d. More Expressive KR for NL understanding (Any compromise?)
    e. Any work exploring how existing representations fall short of
addressing some problems involved in modelling, manipulating or reasoning
(whether reasoning as used to get an interpretation for a certain utterance,
exchange of utterances or what utterances follow from other utterances) with
NL documents
    f. Representations that show how classical logics are not as efficient,
transparent, expressive or where a one-step application of an inference rule
require more (complex) steps in a classical environment and vice-versa; i.e.
how classical logics are more powerful, etc
    g. Building a reasoning test collection for natural language
understanding systems: any kind of reasoning (deductive, abductive, etc);
for a deductive test suite see for e.g. deliverable 16 of the FraCas project
(http://www.cogsci.ed.ac.uk/~fracas/). Also, look at textual entailment
challenges 1 and 2
    h. Comparative results (on a common test suite or a common task) of
different representations or systems that reason with NL (again any kind of
reasoning). The comparison could be either for efficiency, transparency or
    i. Knowledge acquisition systems or techniques that benefit from
properties of NL to acquire knowledge already 'coded' in NL
    j. Automated Reasoning, Theorem Proving and KR communities views on all
Received on Sat Oct 08 2005 - 03:37:25 EDT

This archive was generated by hypermail 2.2.0 : Sat Oct 08 2005 - 03:37:27 EDT