Computational Lexical Semantics of Verbs
April 28 and 29
U. of Pennsylvania
Philadelphia, Pennsylvania, USA
sponsored by the Institute for Research in Cognitive Science
at the U. of Pennsylvania
and IFIP
PROGRAM COMMITTEE: Martha Palmer, Harry Bunt, Bonnie Dorr,
Paul Jacobs, Sergei Nirenburg, James Pustejovsky, Patrick St.
Dizier, Rich Thomason
We would like to hold a two-day workshop on issues in
computational lexical semantics. Long recognized as a critical
component of any natural language system, this area has
paradoxically offered tantalizing glimpses of a wealth of data
for resolving parsing and reference issues, while at the same
time successfully eluded systematic representation. Current
implementations each have their own representation schemas, and
require customized lexical semantic representations, with the
notion of reuse and recycle with respect to lexicons seemingly
quite far out of reach. However, a recent spate of workshops
on dictionaries and on lexical semantics, current developments
in linguistics such as Wordnet [Miller, 1991], and Levin's verb
classes [Levin,1993] , as well as the wide-spread use of MRDs,
(Machine Readable Dictionaries), suggest that the time is ripe
for a push towards commonality.
The aim of this workshop is to bring together influential
researchers in linguistics, text analysis, machine translation,
lexical semantics, formal semantics and knowledge representation
for an in-depth discussion of fundamental issues. The focus
of the workshop will be the discussion of the representational
needs of pre-selected controversial lexical items, all verbs.
Having established these needs, the ability of individual
system implementations to meet these needs will be compared and
contrasted. The utility of data structures such as multi-lingual
ontologies and cross-linguistic verb classification schemes will
be stressed, with an open discussion of possible techniques and
methodologies for determining such data structures. The short
term benefit would be a greater consensus on representation that
would allow researchers to exchange lexicons and morphological
analyzers, and to collaborate on common corpora. The long term
goal is a unified methodology for resolving issues in semantic
representation that would allow the whole to be built from the
sum of its parts - that would encourage the joint development of
a shared core lexicon. It is not expected that a single workshop
can accomplish this objective, but rather that open discussion
can allow us to recognize unexpected areas of agreement and
isolate areas of disagreement. The two day workshop would begin
each day with presentations and panel discussions, and break up
into working groups in the afternoons. These working groups
will examine the pre-selected examples and work through the
comparisons of the implementations of these examples with respect
to different systems. It will be necessary to keep the number
of participants in the workshop quite low, and we will expect the
participants to engage in a certain amount of advance preparation
in order to make the working groups as effective as possible.
The two days will be organized as follows: The first day will
concentrate on semantic problems in text analysis of English,
with presentations of different system approaches. The speakers
will be asked to focus on selected examples, and give details
about their system's treatment of the following issues:
- Interaction between syntax and semantics; definition of
linking rules (by class, by thematic role, by verb argument,
etc.)
- Relation between alternations and verb classes; properties
inherited by verb classes
- Status of thematic roles vs. thematic relations vs. verb
classes
In addition to system approach presentations, we would like to
have three invited speakers, on Knowledge Representation, Lexical
Semantics and Verb Classifications.
The second day the emphasis will shift more to multilingual
analysis, and extending some of the techniques discussed the
day before to other languages. The presentations of different
approaches will look at some of the same examples as the day
before, as well as additional ones, focusing on:
- language-specific primitives vs. interlingua primitives
- language-specific ontologies vs. an interlingua ontology
We would hope to have one more invited talk on Wordnet.
Anyone interested in participating in this workshop is invited to
send a 1/2 page Statement of Interest to mpalmer@linc.cis.upenn.edu
or
Martha Palmer
CIS Department
Moore School
U of Pennsyslvania
Philadelphia, PA 19104-6389
by Feb 17. Participants will be notified by March 17.