THE LOCKHEED MARTIN TIPSTER II PROJECT 
Steven Maiorano 
Office of Research & Development 
Washington, D.C. 20505 
E-mail: smaioran@ord.gov 
Telephone: 703-351-2701 
Introduction 
The Lockheed Martin TIPSTER II project 
focused on several research areas. In addition, three 
demonstration projects are underway to exhibit the 
feasibility of transitioning the research into an 
operational setting. 
in text together with a set of tools for easily 
constructing recognizers for new objects. In addition, 
the group has advanced the state of the art for 
identifying co-referential noun phrases. This work 
was evaluated during MUC-6 and performed extremely 
well. The ability to resolve co-references provides a 
sound basis for all forms of link analysis \[1\] 
TIPSTER II Research Areas 
Lockheed Martin's research accomplishments 
were threefold: 
1. They defined a methodology for constructing 
domain-specific extraction systems in a cost-effective, 
reliable way. The MUC-6 evaluation system for the 
scenario template was constructed in three person 
weeks from the genetic system and performed very 
well. 
2. They provided insight into the benefit of 
particular techniques for integrating extraction and 
detection technology through a series of experimental 
results. The best of these experimental techniques 
will be integrated in the joint Lockheed Martin/GE 
CR&D/NYU/Rutgers submissions for TREC-5. 
3. They contributed to defining a TIPSTER 
architecture, testing it, and implementing a multi- 
platform version of the document manager defined by 
the architecture. The document manager specifies a 
set of functions that should be used to get text for 
text processing and to store the output of text 
processing. Conforming to the TIPSTER architecture 
allows various extraction and detection modules to 
work together. 
The Generic Extraction System 
To accomplish the goal of providing a core extraction 
capability that can be easily customized to a particular 
domain, Lockheed Martin developed a genetic text 
processing system. The genetic system incorporates 
their research in identifying a range of basic entities 
I This material has been reviewed by the CIA. That review neither \[ constitutes CIA authentication of information nor implies CIA I 
endorsement of the author's views. \[ 
The Integration of Detection and 
Extraction 
A range of experiments were performed here to 
integrate extraction capabilities into the SMART 
retrieval system. This work was performed jointly 
with GE CR&D \[2\] and shows promising results for 
large-scale integration; The work will be integrated 
for the TREC-5 evaluation. 
In addition, a range of experiments were 
performed to extend previous classification results \[3\] 
to routing, and to compare the resulting system with 
those used in TREC-4. The comparison will be 
described in the full paper and the results will be 
incorporated into a submission for the TREC-5 
routing systems \[4\] 
The Architecture Working Group 
Lockheed Martin has participated in the 
Architecture Working Group (AWG), and provided a 
testbed of the architecture functions being defined 
throughout the course of the contract. Lockheed 
Martin has implemented a document manager that 
accesses documents stored in a variety of databases. 
They have both a PC and Sun version at this time, 
and currently access information stored in either 
Sybase or Oracle. The NMSU document manager was 
also integrated into the Lockheed Martin text 
processing system (NLToolset) in a flexible way that 
will allow other document managers to be included 
easily. 
47 
The Demonstration Projects Summary 
ADEPT 
The Advanced Data Extraction and Preparation via 
Tipster (ADEPT) Program is a demonstration project 
aimed at alleviating problems currently being faced by 
the Office of Information Resources (OIR). OIR has 
been chartered to implement enterprise-wide systems 
to collect, archive, distribute, and manage information 
obtained from unclassified data sources typically 
called "Open Sources" \[5\]. 
CANIS 
The Cable Abstracting and Indexing System Program 
is aimed at providing assistance to government 
analysts in indexing and abstracting their cable traffic. 
The goal of the project is to reduce the amount of 
manual time that an analyst will spend reviewing and 
processing each cable \[6\] 
FTM 
The Free Text Management project is a 
demonstration system that will be located in the 
Federal Intelligent Document Understanding 
Laboratory (FIDUL); the project will include 
TIPSTER detection and extraction tools integrated 
with a common document manager. 
Overall, Lockheed Martin has demonstrated 
improvements in the technology to process texts and 
to do extraction, the ability to move the research into 
a realistic setting, and the ability to develop systems 
conforming to the architecture. 

References 

\[1\] Childs, L.; "An Evaluation of Coreference 
Resolution Strategies for Acquiring Associated 
Information;" in this volume. Tipster 1996

\[2\] Guthrie, L., Strzalkowski, T., Fang Lin, and 
Jin Wang; "Integration of Document Detection and 
Information Extraction;" in this volume. Tipster 1996

\[3\] Guthrie, L., Walker, E., and Guthrie, J.; 
"Document Classification by Machine: Theory and 
Practice;" in Proceedings of the 16th International 
Conference on Computational Linguistics (COLING 
.~.); Kyoto, Japan; 1059-1063; 1994. 

\[4\] Guthrie, L. and Leistensnider, J.; "A Simple 
Probalistic Approach to Classification and Routing;" 
in this volume. Tipster 1996

\[5\] Kielty, J., and Sider, I.; "Advanced Data 
Extraction and Preparation via Tipster (ADEPT);" in 
this volume. Tipster 1996

\[6\] Sider, I., Baker, J., Higbie, L., and Howard, 
T.; "Cable Abstracting and Indexing System ( CANIS) 
Prototype;" in this volume Tipster 1996
