<?xml version="1.0" standalone="yes"?>
<Paper uid="H86-1008">
  <Title>A Terminological Simplification Transformation for Natural Language Question-Answering Systems 1</Title>
  <Section position="19" start_page="2238" end_page="2238" type="concl">
    <SectionTitle>
5. Conclusions, Implementation Status and Further
</SectionTitle>
    <Paragraph position="0"> Work As of this writing, we have incorporated NIKL into the implementation of our natural language question-answering system, IRUS. NIKL is used to represent the knowledge in a Navy battle-management domain. The simplification transformation described in this paper has been implemented in this combined system, and the axiomatization of the .database as described above is being added to the domain model. At that point, the  methodology will be tested as a solution to the difficulties now being experienced by those trying to write the translation rules for the complex database and domain of the Fleet Command Center .Battle Management Program of DARPA's Strategic Computing Program.</Paragraph>
    <Paragraph position="1"> I have presented a limited inference method on predicate calculus expressions, whose intent is to place them in a canonical form that makes other inferences easier to make. Metaphorically, it can be regarded as &amp;quot;sinking&amp;quot; the expression lower in a certain logical space. The goal is to push it down to the &amp;quot;level&amp;quot; of the database predicates, or below. We cannot guarantee that we will always place the expression as low as it could possibly go - that problem is undecidable. But we can go a good distance, and this by itself is very useful for restoring the tractability of the mapping transformation and other sorts of deductive operations \[10\].</Paragraph>
    <Paragraph position="2"> Somewhat similar simplifications are performed in the work on ARGON \[6\], but for a different purpose. There the database is assumed to be a full, rather than a partially specified, model and simplifications are performed only to gain an increase in efficiency. The distinguishing feature of the present work is its operation on an expression in a logical language for English meaning representation, rather than for restricted queries. A database, given the purposes for which it is designed, cannot constitute a full model for such a language. Thus, the terminological simplification is needed to reduce the logical expression, when possible, to an expression in a &amp;quot;sublanguage&amp;quot; of the first for which the database is a full model.</Paragraph>
    <Paragraph position="3"> An importar~t outcome of this work is the perspective it gives on knowledge representation systems like NIKL. It shows how workers in other fields, while maintaining other logical systems as their primary mode of representation, can use these systems in practical ways. Certainly NIKL and NIKL-like systems could never be used as full meaning representations - they don't have enough expressive power, and were never meant to. This does not mean we have to disregard them, however. The right perspective is to view them as attached inference engines to perform limited tasks having to do with their specialty - the relationships between the various properties and relations that make up a subject domain in the real world.</Paragraph>
  </Section>
class="xml-element"></Paper>