HLTCentral HLTCentral HLTCentral
HLTCentral - the gateway to speech and language technology opportunities  
 
Pathway to market for speech and language technologies
European network of excellence in HLT
EC Programme fostering digital content on the global networks
HLTCentralHomeInternational news items from the pressAbout HLTSitemapSearch
HLTCentral    
 About HLTCentral
 Events
 FP6 - Way Forward
 Calls
 Links
 Repository
 What's new
 Who's Who
 >Projects
   Alphabetical List
   Programme List
   Group List
   Search Projects
DIET DIET
 
Summary Results Consortium Index

Diagnostic and Evaluation Tools for Natural language Applications - DIET

The DiET project has developed a comprehensive environment for the construction, annotation and maintenance of structured reference data for the diagnosis and evaluation of NLP (Natural Language Processing) applications for developers and professional evaluators. The system is based on an open client/server architecture and allows test data to be annotated by choosing from annotation types, and comes with edit, display and storage functions. The project also produced structured test data on the morphology, syntax and discourse levels. The user can adapt the data to new domains through customisation functions. Links can be established between the controlled test items and related phenomena in domain specific corpora through text profiling. Lexical replacement allows users to adapt the vocabulary to their specific terminology. The tools support the set-up of evaluation scenarios and the recording of test results.

Objectives

The main goal is to develop the methods and tools for the glass box evaluation of NL components. This includes:

  • checking the performance of an NL system against well defined linguistic phenomena identified in real corpora, such as maintenance manuals;
  • measuring the evolution of a system under development through different releases to identify improvements and detect possible regressions;
  • enhancing resources developed in previous projects such as annotation schema and test suites;
  • introducing techniques which use morphology, semantics and discourse as well as tools for domain and corpus based customisation, including the tailoring to specific sub-lexicons;
  • testing the syntactic and semantic competence of an NL system.

The technology used involves the construction, annotation and application of systematic NL test suites. Database and evaluation technologies, as well as statistical and corpus annotation methods, are based on the results of the EU funded projects TSNLP, FRACAS, TEMAA, EAGLES, and of the commercially funded SLT (Spoken Language Translator) project.

Benefits & Users

Effective and efficient assessment of Natural Language (NL) components is often severely hampered by the lack of suitable test material and technology, which are expensive and time consuming to develop. Available evaluation tools are highly specialised and unsuitable for reuse. DIET provides a set of reusable diagnostic and evaluation tools and data needed by suppliers to monitor the quality of NLP products and services.

The users in the consortium have developed different evaluation scenarios. AEROSPATIALE already has significant experience in using and evaluating textual information processing systems for document production and manipulation where it needs to manage and produce huge amounts of documentation in several different languages. Aerospatiale scenarios cover two applications - controlled language checking and grammar checking.

The Localisation Resources Centre evaluates tools and assists clients with the implementation of suitable tools by providing a consultancy and training service. LRCs scenarios focus on translation memory systems, which are often used in the localisation process, and investigates the need for new data and annotation types.

As a major technology provider, IBM is a prime user of NLP test suites and tools. A systematic test suite helps to overcome inherent limits of ad hoc compiled test data and a ready-made and flexible diagnostics and evaluation tool facilitates quality assurance. The scenario at IBM focuses on the use of the lexical replacement module to generate derived test suites.

Even though the benefits have been demonstrated in specific evaluation scenarios, the flexibility and adaptability of the tools allows for a much broader range of applications. The results could be exploited for any kind of construction and maintenance of structured linguistic resources and interest has already been expressed by a number of external users.

Go to the DIET web site

Print  Project factsheet optimized for printing Last updated: 12.01.2001

[  Projects Home  |  Alphabetical List  |  Programme List  |  Group List  |  Search Projects  ]


Please report problems to