<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-3025">
  <Title>Is there content in empty heads?*</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> In order to extract meaning from text, we must at least disambiguate the words used in the text. The normal sceaario is to locate the disambiguated words in some knowledge base, which gives additional infc, m~,ation about the words and their properties, so that some of the higher-minded tasks of natural i;mgt.,age processing (NLP) can be accomplished: tasks iike determining speech acts, identifying topic a~_d focus shifts for discourse analysis, and drawing inferences. The work described here is a step reward developing an initiN knowledge base for NLP by automatmally transforming information found in machine readable dictionaries into a data base suitable for a variety of NLP applications.</Paragraph>
    <Paragraph position="1"> The overall scheme of the work at CRL on Inachine readable dictionaries is described in Wilks et al. (1988, 1989, 1990). As part of that work, Slator (1988a, 1988b; Slator and Wilks, 1987, 1990) developed a program called Lexicon Provider which creates frames from the dictionary definitions of word senses provided by the Longman's Dictionary of Contemporary English (LDOCE). Each frame is con** Prc~cnt address: Department of Computer Science, North Dakota State University, Fargo, ND 58105 nected via an IS-A link to some other word (spelling form) in the dictionary. Our goal in this paper is to refine that work so as to connect each frame to another word sense in the dictionary. This insures that properties can be consistently inherited in this graph structure (since A IS-A B allows A to inherit properties of B). We can think of this task as building a taxonomy of the word senses in LDOCE.</Paragraph>
    <Paragraph position="2"> This paper presents our techniques for automating this task for noun definitions, using the special c~xled information found in the machine readable version of LDOCE. We also present ways to extract other semantic relations automatically as part of the process.</Paragraph>
  </Section>
class="xml-element"></Paper>