Project Penlight - A Government Perspective 
Michael J. Chrzanowski 
Department of Defense 
9800 Savage Road, Fort Meade, MD 20755 
mjchrza@ romulus.ncsc.mil 
(301) 688-9131 
The purpose of the TIPSTER contract with the Uni- 
versity of Pennsylavnia (Project PENLIGHT) was to 
promote and further develop Information Extraction and 
Text Summarization by the application of coreference 
resolution, an integral part of IE systems, to Text Sum- 
marization. 
During the first two years of this effort, which 
began on 1 October 1996, the University of Pennsylva- 
nia was charged with two parallel tasks. The greater 
part of its effort was to go into the design and implemen- 
tation of the coreference engine. The second task, 
essentially an independent effort, was to automatically 
detect discourse segments. 
The University of Pennsylvania has been very suc- 
cessful in developing the coreference engine. The work 
in this area has been recognized as valuable by other 
researchers in the Summarization effort. As a result, the 
University will be making its coreference software 
(known as CAMP) available to the research community 
on the World Wide Web. The potential importance of 
this work as a whole to the automatic text surnmariza- 
tion effort has caused the Government to slighly re-align 
its priorities over the course of the first two contract 
years. The Government instructed the University of 
Pennsylvania to spend almost all of its efforts in improv- 
ing the coreference resolution engine and making it 
available to the research community rather than apply- 
ing this engine to the problems of text summarization. 
While the results of the coreference resolution 
development work have been impressive, the re,suits of 
the application of this effort to text summarization have 
also achieved success, particularly in the area of query- 
specific summarization. The University of Pennsylvania 
approach is to resolve all coreferences within the full 
text to the query and select all sentences which are 
coreferent. Results from the dry-run and SUMMAC 
evaluations imply that coreference resolution is a neces- 
sary element of good summarization. 
The University of Pennsylvania continues its efforts 
in resolving cross-document coreference which will be 
necessary to perform multi-document summarization. 
