Workshop on Distributing and Accessing Linguistic Resources

Granada, Spain - May 27 1998 (afternoon)

Short description

This workshop will discuss ways to increase the efficacy of linguistic resource distribution and programmatic access, and work towards the definition of a new method for these tasks based on distributed processing and object-oriented modelling with deployment on the WWW.

Organizers: Yorick Wilks, Hamish Cunningham, Wim Peters, Remi Zajac


The following papers will be presented in order of enumeration. After each 15 minute presentation there will be 5 minutes for discussion.

Distributed Thesaurus Storage and Access in a Cultural Domain Application
S. Boutsis, B. Georgantopoulos, S. Piperidis
Institute for Language and Speech Processing, Athens

A New Model for Language Resource Access and Distribution
W. Peters, H. Cunningham, Y. Wilks, C. McCauley
University of Sheffield

Reuse and Integration of NLP Components in the Calypso Architecture
R. Zajac
New Mexico State University

Corpus-based Research using the Internet
H. Brugman, A. Russel, P. Wittenburg
Max Planck Institute for Psycholinguistics, Nijmegen

The CUE Corpus Access Tool
O. Mason
University of Birmingham

Linguistic Research Utilizing the EDR Electronic Dictionary as a Linguistic Resource
T. Ogino
EDR, Japan


The following posters will be on display during the workshop, and presentations are planned during the breaks:

TRACTOR: TELRI Research Archive of Computational Tools and Resources
R. Krishnamurthy
University of Birmingham

Web-Surfing the Lexicon
D. Cabrero, M. Vilares, L. Docampo, S. Sotelo
Ramon Pineiro Research Centre/Universities of Coruna and Santiago

Exploring Distributed MT
O. Streiter, A. Schmidt-Wigger, U. Reuther, C. Pease
IAI Saarbruecken

A Proposal for an On-line Lexical Database
P. Cassidy
Micra, Inc.


The final part of the workshop will consist of a panel discussion on:

Distributing and Accessing Linguistic Resources

The panel participants are:
Khalid Choukri, Eduard Hovy, Judith Klavans, Yorick Wilks, and Antonio Zampolli.


In general the reuse of of NLP data resources (such as lexicons or corpora) has exceeded that of algorithmic resources (such as lemmatisers or parsers). However, there are still two barriers to data resource reuse:

  1. each resource has its own representation syntax and corresponding programmatic access mode (e.g. SQL for CELEX, C or Prolog for Wordnet, SGML for the BNC);
  2. resources must generally be installed locally to be usable (and of course precisely how this happens, what operating systems are supported etc. varies from case to case).
The consequences of 1) are that although resources share some structure in common (lexicons are organised around words, for example) this commonality is wasted when it comes to using a new resource (the developer has to learn everything afresh each time) and that work which seeks to investigate or exploit commonalities between resources (e.g. to link several lexicons to an ontology) has to first build a layer of access routines on top of each resources. So, for example, if we wish to do task-based evaluation of lexicons by measuring the relative performance of an information extraction system with different instantiations of lexical resource, we might end up writing code to translate several different resources into SQL or SGML.

The consequence of 2) is that there is no way to "try before you buy": no way to examine a data resource for its suitability for your needs before licencing it. Correspondingly there is no way for a resource provider to expose limitted access to their products for advertising purposes, or gain revenue through piecemeal supply of sections of a resource.

This workshop will discuss ways to overcome these barriers. The proposers will discuss a new method for distributing and accessing language resources involving the development of a common programmatic model of the various resources types, implemented in CORBA IDL and/or Java, along with a distributed server for non-local access. This model is being designed as part of the GATE project (General Architecture for Text Engineering) and goes under the provisional title of an Active CREOLE Server. (CREOLE: Collection of REusable Objects for Language Engineering. Currently CREOLE supports only algortihmic objects, but will be extended to data objects.)

A common model of language data resources would be a set of inheritance hierarchies making up a forest or set of graphs. At the top of the hierarchies would be very general abstractions from resources (e.g. lexicons are about words); at the leaves would be data items that were specific to individual resources. Programmatic access would be available at all levels, allowing the developer to select an appropriate level of commonality for each application.

Note that although an exciting element of the work could be to provide algorithms to dynamically merge common resources (e.g. connect WordNet to Celex), what we're suggesting initially is not to develop anything substantively new, but simply to improve access to existing resources. This is NOT a new standards initiative, but a way to build on previous initiatives.

Of course, the production of a common model that fully expressed all the subtleties of all resources would be a large undertaking, but we believe that it can be done incrementally, with useful results at each stage. Early versions will stop decomposing the object structure of resources at a fairly high level, leaving the developer to handle the data structures native to the resources at the leaves of the forest. There should still be a substantial benefit in uniform access to higher level strucures.

Draft Program Committee

Yorick Wilks
Hamish Cunningham
Wim Peters
Remi Zajac
Roberta Catizone
Paola Velardi
Maria Teresa Pazienza
Louise Guthrie
Roberto Basili
Bran Boguraev
Sergei Nirenburg
James Pustejowsky
Ralph Grishman
Christiane Fellbaum