<?xml version="1.0" standalone="yes"?>
<Paper uid="J87-3005">
  <Title>DISAMBIGUATING PREPOSITIONAL PHRASE ATTACHMENTS ON-LINE DICTIONARY DEFINITIONS BY USING</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1. INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> A customary pattern for papers on natural language processing runs roughly as follows:  1. Here's a difficult language situation. 2. Here's the semantic (pragmatic / discourse / whatever) information necessary to interpret the situation as we humans interpret it.</Paragraph>
    <Paragraph position="1"> 3. Here's how I put this information into my system. 4. Therefore, my system can handle the situation.  Although the amount of effort which has been expended in doing this kind of thing is admirable, there is still something less than totally satisfactory about the approach. First, it tends to trivialize the notion of &amp;quot;solution to a problem.&amp;quot; Second, and more important, the approach relies on hand-coding pieces of information (commonly called &amp;quot;knowledge&amp;quot;). The more we want our programs to know, the more we have to hand-feed them. We can keep on doing this, and maybe, after several generations, someone will put all the pieces together and discover that AI researchers have succeeded in rewriting the history of the universe as we understand it.</Paragraph>
    <Paragraph position="2"> Without question, this will be a form of success. But there might be a faster way to the goal ---one that doesn't involve reinventing the wheel of knowledge. We * The second author currently works for B.I.M., Belgium. already have volumes of hand-coded natural language information in our reference books ---dictionaries and encyclopedias, for instance. If computers could access that information, and use it to help get out of difficult situations, they (and we) would be much further ahead. Problems should be expected, of course: reference works are arbitrary and inconsistent; just having the information does not guarantee being able to use it.</Paragraph>
    <Paragraph position="3"> It is nevertheless possible to think of on-line reference books as knowledge bases. In this paper we propose techniques for processing the definitions of an on-line standard dictionary, and for extracting from them the semantic information necessary to resolve the ambiguities inherent in the attachment of English prepositional phrases. We consult the on-line dictionary as if it were a semantic expert, and we find in it the kind of information that has previously been supplied by means of scripts, frames, templates, and similar hand-crafted devices.</Paragraph>
    <Paragraph position="4"> We start by presenting PEG, a broad-coverage computational grammar of English which is our tool for analyzing on-line definitions, and by discussing the tools necessary for extracting from these definitions the knowledge relevant for disambiguation. Subsequent sections describe these tools in greater detail, focussing on specific sentences and on the heuristic machinery used to discover their most likely interpretations. A final section sums up the work and suggests some Copyright 1987 by the Association for Computational Linguistics. Permission to copy without fee all or part of this material is granted provided that the copies are not made for direct commercial advantage and the CL reference and this copyright notice are included on the first page. To copy otherwise, or to republish, requires a fee and/or specific permission. 0362-613X/87/030251-260503.00 Computational Linguistics Volume 13, Numbers 3-4, July-December 1987 251 Karen Jensen and Jean-Louis Binot Disambiguating Prepositional Phrase Attachment important areas for future research. Two appendices provide traces of the processing of some examples, and technical information about the &amp;quot;approximate reasoning&amp;quot; techniques used in our system.</Paragraph>
  </Section>
class="xml-element"></Paper>