<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-1107">
  <Title>REXTOR: A System for Generating Relations from Natural Language</Title>
  <Section position="3" start_page="67" end_page="67" type="intro">
    <SectionTitle>
2 Motivation
</SectionTitle>
    <Paragraph position="0"> We believe that, for humans, natural language is the best mechanism for information access. It is intuitive, easy to use, rapidly deployable, and requires no specialized training, The REXTOR System builds on the experience of START (SynTactic Analysis using Reversible Transformations), a natural language system available for question answering on the World Wide Web. 2 Since December, 1993, when it first 2hZZp :/\[~ww. ai .mit. edu/projects/infolab came online, START has engaged in millions of exchanges with hundreds of thousands of people all over the world, supplying users with knowledge regarding geography, weather, movies, and many many other areas. Despite the successes of START in serving actual users, its domain of knowledge is relatively small and expanding its knowledge base is a time-consuming task. The goal of REXTOR is to overcome this bottleneck and to provide a general framework for natural-language information retrieval. REXTOR not only draws its inspiration from START (in providing question answering capabilities), but also borrows a simplified form of its representational structures (Katz, 1980; Katz, 1990).</Paragraph>
    <Paragraph position="1"> The START System (Katz, 1990; Katz, 1997) analyzes English text and builds a knowledge base from information found in the text. The knowledge is expressed in the form of embedded ternary expressions (T-expressions) -- subject-relation-object triples where the subject and object can themselves be ternary expressions. For example, &amp;quot;The population of Zimbabwe is 11,044,147&amp;quot; would be represented as two ternary expressions:</Paragraph>
  </Section>
class="xml-element"></Paper>