<?xml version="1.0" standalone="yes"?>
<Paper uid="M91-1033">
  <Title>UNIVERSITY OF MASSACHUSETTS: DESCRIPTION OF THE CIRCUS SYSTEM AS USED FOR MUC-3</Title>
  <Section position="2" start_page="0" end_page="223" type="intro">
    <SectionTitle>
BACKGROUND AND MOTIVATION
</SectionTitle>
    <Paragraph position="0"> In 1988 Professor Wendy Lehnert completed the initial implementation of a semantically-oriente d sentence analyzer named CIRCUS [1] . The original design for CIRCUS was motivated by two basic research interests : (1) we wanted to increase the level of syntactic sophistication associated wit h semantically-oriented parsers, and (2) we wanted to integrate traditional symbolic techniques i n natural language processing with connectionist techniques in an effort to exploit the complementar y strengths of these two computational paradigms .</Paragraph>
    <Paragraph position="1"> Shortly thereafter, two graduate students, Claire Cardie and Ellen Riloff, began to experimen t with CIRCUS as a mechanism for analyzing citation sentences in the scientific literature [2] . The key idea behind this work was to extract a relatively abstract level of information from each sentence , using only a limited vocabulary that was hand-crafted to handle a restricted set of target concepts . We called this mode of language processing selective concept extraction, and the basic style of sentence analysis was a type of text skimming . This project provided us with an opportunity to give CIRCUS a workout and determine whether or not the basic design was working as expected .</Paragraph>
    <Paragraph position="2"> Although CIRCUS was subject to a number of limitations, the integration of syntax and semantic s appeared to work very nicely . We believed we had constructed a robust text skimmer that wa s semantically oriented but nevertheless able to use syntactic knowledge as needed . Projects associated with the connectionist aspect of CIRCUS took off at about this time and carried us in those directions for a while [3,4,5] . When an announcement for MUC-3 reached us in June of 1990, we felt that the MUC- 3 evaluation required selective concept extraction capabilities of just the sort we had been developing .</Paragraph>
    <Paragraph position="3"> We were eager to put CIRCUS to the test.</Paragraph>
    <Paragraph position="4"> It was clear to us that MUC-3 would require a much more ambitious and demanding application o f CIRCUS than our earlier work on citation sentences, and we fully expected to learn a great deal fro m the experience. We hoped to capitalize on Cardie and Riloff's previous experience with CIRCUS whil e identifying some new areas for ongoing research in sophisticated text analysis . In September of 1990, Robert Williams joined our MUC-3 effort as a post doc with research experience in case-base d reasoning. Cardie, Riloff, and Williams provided the technical muscle for all of our MUC-3 syste m development and knowledge engineering. Cardie was primarily responsible for CIRCUS and dictionar y design, Riloff developed the rule-based consolidation component, and Williams designed the case based reasoning consolidation component. Although the division of labor was fairly clean, everyone worked with CIRCUS dictionary definitions and the preprocessor at various times as needed . In January of 1991, David Fisher joined the project as an undergraduate assistant who designed a n interface for faster dictionary development while assisting with internal testing . Professor Lehnert assumed a leadership role but made no programming contributions to MUC-3.</Paragraph>
  </Section>
class="xml-element"></Paper>