NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/comp/ai/3303

Internet Message Format | 1992-09-02 | 6.5 KB

Path: sparky!uunet!cis.ohio-state.edu!ucbvax!UMIACS.UMD.EDU!bonnie
From: [e-mail] (Bonnie J. Dorr)
Newsgroups: comp.ai
Subject: CALL FOR PARTICIPATION: AAAI-93 SPRING SYMPOSIUM SERIES
Message-ID: <[e-mail]>
Date: 2 Sep 92 15:39:42 GMT
Sender: [e-mail]
Lines: 125

[I apologize for sending this message again.  I left out the header last time.]

CALL FOR PARTICIPATION: AAAI-93 SPRING SYMPOSIUM SERIES

              BUILDING LEXICONS FOR MACHINE TRANSLATION

                AAAI-93 Spring Symposium Series

                    March 23--25, 1993

                  Stanford, University, CA

The lexicon plays a central role in any machine translation (MT)
system, regardless of the theoretical foundations upon which the
system is based.  However, it is only recently that MT researchers
have begun to focus more specifically on issues that concern the
lexicon, e.g., the automatic construction of multi-lingual semantic
representations.  Large dictionaries are important in any natural
language application, but the problem is especially difficult for MT
because of cross-linguistic divergences and mismatches that arise from
the perspective of the lexicon.  Furthermore, scaling up dictionaries
is an essential requirement for MT that can no longer be dismissed.

This symposium provides a forum for researchers from the fields of MT
and the lexicon focus on the intersection of the two fields, rather
than their broader concerns.  A number of fundamental questions will
be addressed:

-- What lexical levels are required by a machine translation
   system?  Syntactic? Lexical semantic? Ontological?  What do the
   representations at each of these levels look like, and how
   would they be constructed?

-- What are the interdependencies between these levels?  Can we
   take advantage of interacting linguistic constraints from each
   level for the construction of lexical representations?  Should
   the levels be kept as separate layers and related explicitly
   or should they be combined into one layer and be related
   implicitly?  Should all levels be represented in the same or
   in different, dedicated formalisms? What are the implications of these
   choices for MT system architecture, processing of the relevant
   knowledge, interaction between components of MT systems, applicability
   of the resulting knowledge sources in different types of MT mappings?

-- Can automatic procedures be used for the construction of
   lexical representations?  What existing resources should we be using
   and what aids do we have to transform these resources into appropriate
   representations for MT?  To what extent is it possible to acquire
   elements of contrastive knowledge (mapping information) using existing
   techniques (e.g., work on bilingual corpora, example based approaches,
   etc.)?

-- To what extent is it possible to share lexicons?  If the
   representations and the actual knowledge are tailored to
   a specific system (e.g., style of grammar or choice of domain
   knowledge base) then how can sharing be achieved?
   How much representations and knowledge are tied to specific
   approaches to MT system construction, and, to the extent that
   they are, how much can people come to some agreement on some of
   those other issues so that they can share lexicons?

-- Are bilingual dictionaries useful for the construction of
   computational lexicons for MT?  What is the role of example
   sentences and phrases in bilingual dictionaries?  Can we extract
   information from pairwise examples in order to achieve example-based
   translation?  Can we use bilingual dictionaries for the extraction of
   grammatical information?

-- What are the different types of MT mappings (transfer,
   interlingual, statistically based, memory-based, etc.) and how do
   these mappings affect the representation that is used in the lexicon?

-- What types of MT divergences and mismatches must be accommodated
   in the lexicon (i.e., cases where the target-language sentence
   has a different structure, or conveys different information,
   from that of the source language)? Are these problems that
   any translation system must deal with regardless of the MT mapping
   that is used?  If so, can we construct lexicons that accommodate these
   divergences regardless of the translation mapping that is used?  Can
   we incorporate information about the respective portions of
   lexical/non-lexical knowledge needed to decide on suitable candidates
   for target constructions and on lexical clues for strategies for such
   decisions?

-- Can we, or have we, achieved language-independence in the
   representations that are used in the lexicon?  Can we support an
   interlingual approach to machine translation based on current
   technology and resources?

All interested participants should submit five copies of a one- to
five-page abstract (not including the bibliography) by October 16,
1992 to:

Bonnie Dorr
Department of Computer Science / UMIACS
University of Maryland
A.V. Williams Building
College Park, MD 20742

FAX or electronic submission will not be accepted.  Each submission
should include the names and complete addresses of all authors.
Correspondence will be sent to the authors by e-mail, unless otherwise
indicated.  Also, authors should indicate under the title which of the
questions and/or topic listed above best describes their paper (if
none is appropriate, please give a set of keywords that best describe
the topic of the paper).

Authors will be notified of the Program Committee's decision by
November 16, 1992.  Submissions will be judged on clarity,
significance, and originality.  An important criterion for acceptance
is that the abstract clearly contributes to the theme of building
lexicons for machine translation.  Abstracts focusing on one of these
two areas (i.e., MT or the lexicon) will be given a lower priority
than those that address issues that lie at their intersection.

Program Committee: Michael Brent ([e-mail]), Johns
Hopkins University; Bonnie Dorr (chair) ([e-mail]),
University of Maryland; Sergei Nirenburg ([e-mail]),
Carnegie Mellon University; Elaine Rich ([e-mail]),
Microelectronics and Computer Technology; Patrick Saint-Dizier
([e-mail]), CNRS, Universite' Paul Sabatier

[Note: Registration information will be available in December.           ]
[      To obtain registration information (including cost) write to the  ]
[      AAAI at 445 Burgess Drive, Menlo Park, CA 94025 ([e-mail])    ]