TACP logo

National Institute of Standards and Technology Home Page
TIPSTER Text Program A multi-agency, multi-contractor program


TIPSTER Overview
TIPSTER Technology Overview
TIPSTER Related Research
Phase III Overview
TIPSTER Calendar
Reinvention Laboratory Project
What's New

Conceptual Papers
Generic Information Retrieval
Generic Text Extraction
Summarization Concepts
12 Month Workshop Notes

Text Retrieval Conference
Multilingual Entity Task
Summarization Evaluation

More Information
Other Related Projects
Document Down Loading
Request for Change (RFC)
Glossary of Terms
TIPSTER Source Information

Return to Retrieval Group home page
Return to IAD home page

Last updated:

Date created: Monday, 31-Jul-00

Phase III Overview

TIPSTER Phase III - A Four Part Program

DARPA and other TIPSTER members sponsored 17 research and architecture development contracts with academic institutions and commercial companies in their effort to continue a balanced overall program, consisting of four basic parts: Advanced Research, Metrics-based Evaluations, a Structured Software Architecture, and Demonstration and Implementation Projects. Phase III Kick-off workshop was held October 1996.

Advanced Research

TIPSTER's Phase III was aimed at continuing innovative research in basic areas while adding several new topics of investigation: detection (improving search algorithms; merging results from different engines); extraction (tuning systems to new domains; raising accuracy); summarizing (producing a single summary of multiple documents); multilingual capabilities (porting tools and techniques proven in one language to work in other languages); and cross-technology (sharing of information between detection and extraction tools).

Metrics-based Evaluations

The Message Understanding Conferences (MUC) was a forum for assessing and discussing progress in natural language processing which showcased high performing systems in 1995 with scores about 90% in named entity tagging task. The Text Retrieval Conference (TREC), initiated by TIPSTER to evaluate document detection performance, will continue under NIST leadership. Acquisitions boosted the TREC data base to five gigabytes, an invaluable resource for the Information Retrieval (IR) community to test querying and ranking methods. The Summarization Evaluation Conference (SEC) recently completed an initial test of methods to evaluate summarization performance. Some results can be found here.

Software Architecture

Phase III proposed a feature called the TIPSTER Architecture Capabilities Platform (ACP). ACP's goals were aimed at providing framework for research and development in both Document Detection and Information Extraction. This resource, began development in 1997, it focused on providing the community with the opportunity to test components in a TIPSTER Architecture compliant environment and to perform experiments using TIPSTER components and modules. Inclusion of CORBA capabilities and the Z39.50 Information Retrieval Protocol supported reusable components and a more 'standards' base approach to the Architecture.

Demonstration and Implementation Projects

Initial prototyping efforts in Phase II led to several demonstrations of the capabilities of TIPSTER components against operational tasks in the Intelligence Community and elsewhere. The most successful of these early systems was prepared to be migrated to operational use. Phase III proposed a new round of prototyping and applications development. The expansion of the Software Architecture and to bring the results to the user as quickly as possible.

Phase III Work

New Research Areas

Several new research areas were added as Phase III tasks :

  • Text summarization - an enhancement in the information extraction area to develop methods and algorithms to produce, in reasonable English, a summary for each document of interest or a single summary of multiple documents in a collection of interest
  • Merging search results - develop the means to merge results from different search algorithms while maintaining a useful relevance ranking for the retrieved items or fuse the retrieved information with other items
  • Coreference resolution - develop algorithms to resolve text references to the same or different entities; improve the automatic extraction of relationships among entities mentioned in a document

Continuing Research Areas

A number of research areas from Phase I and II required further effort in Phase III :

  • Customization methods - develop more effective ways for a system administrator or end-user to port tools and techniques shown to work in one language or domain to other languages or domains
  • Multilingual capability enhancements in both detection and extraction
  • Further improvements in recall and precision
  • User-interface design and usability testing

Architecture and Capabilities Platform Development

The major focus of the Architecture component for Phase III was the development of a COmmon Request Broker Architecture (CORBA) compliant Architecture Capabilities Platform (ACP) to host TIPSTER-compliant software components and modules. The ACP planned to provide a software platform for testing of individual TIPSTER tools and capabilities. Plans proposed to have developers demonstrate to the Government the modularity of their text handling systems by plugging components and modules into the ACP and interacting with the other TIPSTER components on the platform. In addition, the ACP intended to demonstrate the capability to interact with systems based on Z39.50 standards. Various supporting components for the ACP would have included document collections, standard detection needs, lexicons, a document manager and a default Graphical User Interface (GUI). As a continuation from Phase II, the TIPSTER program was aimming to foster a cooperative effort among the research entities and the ACP developers to provide enhancements to the TIPSTER Architecture design.