<?xml version="1.0" standalone="yes"?>
<Paper uid="P05-3004">
  <Title>CL Research's Knowledge Management System</Title>
  <Section position="4" start_page="0" end_page="13" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> In participating the TREC question-answering track, CL Research began by parsing full documents and developing databases consisting of semantic relation triples (Litkowski, 1999). The database approach proved to be quite confining, with time requirements expanding exponentially trying to maintain larger sets of documents and increasingly complex procedures to answer questions. A suggestion was made to tag text with the type of questions they could answer (e.g., tagging time phrases as answering when questions and person names as answering who questions). This led to the general approach of analyzing parse trees to construct an XML representation of texts (i.e., attaching metadata to the text) and examining these representations with XPath expressions to answer questions.</Paragraph>
    <Paragraph position="1"> Litkowski (2003a) demonstrated the viability of this approach by showing that XPath expressions could be used to answer questions at a level above the highest performing team. Many issues and problems were identified: (1) The necessary level of analysis to meet the needs of particular applications; (2) tagging alternatives; and (3) the viability of the using the XML representation for text summarization, information extraction, novelty detection, and text mining. Subsequent efforts showed that XML representations could be effectively used in summarization (Litkowski, 2003b) and novelty detection (Litkowski, 2005).</Paragraph>
    <Paragraph position="2"> Initially, CL Research developed an interface for examining question-answering performance. This interface has since evolved into a Knowledge Management System (KMS) that provides a single platform for examining English documents (e.g., newswire and research papers) and for generating different types of output (e.g., answers to questions, summaries, and document ontologies), also in XML representations. In this demonstration, CL Research will describe many parts of KMS, particularly the approaches used for analyzing texts.1 The demonstration will particularly focus on the value of XML in providing a flexible and extensible mechanism for implementing the various NLP functionalities. In addition, the demonstration will identify the emerging issue of user modeling to determine exactly how knowledge will be used, since  the primary purpose of KMS is to serve as a tool that will enable users (such as scientists and intelligence analysts) to accumulate and manage knowledge (including facts, such as described in Fiszman et al., 2003) about topics of interest.2</Paragraph>
  </Section>
class="xml-element"></Paper>