-- Bruno Tersago
*------------------------------------------------------------------------*
|Bruno Tersago Tel: +32-16-32 50 88|
|Centre for Computational Linguistics Fax: +32-16-32 50 98|
|Maria-Theresiastraat 21 E-mail: Bruno.Tersago@ccl.kuleuven.ac.be|
|B-3000 Leuven (Belgium) URL: http://www.ccl.kuleuven.ac.be/~bruno/|
*------------------------------------------------------------------------*
CLIF (Computational Linguistics In Flanders) and the Centre for
Computational Linguistics of the Katholieke Universiteit Leuven
have the pleasure of inviting you to the following lecture:
TOWARDS A FRIENDLIER RELATIONSHIP BETWEEN SUBLANGUAGE AND CONTROLLED LANGUAGE
by
RICHARD KITTREDGE
Dept. of Linguistics and Translation, University of Montreal
and
CoGenTex, Inc.
Tuesday 18 June 1996
10:00 - 12:00
Katholieke Universiteit Leuven
Faculty of Arts
Mgr. Sencie-Institute (MSI), room 00.28
Blijde Inkomststraat 21, 3000 Leuven
BELGIUM
Abstract
A certain tension exists between (1) the need of domain experts to
communicate among themselves in an efficient subsystem of their
shared natural language, and (2) the need to communicate technical
information from the experts to "outsiders" in some understandable
way. (The outsiders might be non-native speakers, or have limited
domain knowledge, or both.) A controlled language is designed to
meet the second need under certain assumptions about the domain
and linguistic knowledge of the CL consumers (i.e., the outsiders).
Often, however, little attention is given to the "natural sublanguage"
of the experts, whose knowledge is to be communicated. A closer
mirroring of sublanguage grammatical features during CL design
might help improve acceptance among both CL producer and user
communities.
This talk starts from the notion of sublanguage as it has been
developed and used in North America for information retrieval,
machine translation and (most recently) text generation. We discuss
problems that have arisen or could arise in trying to make a single CL
standard fit a family of sublanguages, or even a whole text genre. We
argue for using better sublanguage models to make the design of CLs
more rational, eventually allowing a more flexible and tailored set of
standards. Current work in natural language generation already faces
issues of whether or how to normalize the output text. Research into
multilingual generation of technical manuals seems to indicate a
future convergence of concerns between the language generation and
CL communities. What the CL community can bring to this
discussion in the short term is the collection and analysis of cases
where CL features have proved problematic to users, or have seems
especially alien to experts.