<?xml version="1.0" standalone="yes"?>
<Paper uid="J86-2001">
  <Title>INTEGRATED PROCESSING PRODUCES ROBUST UNDERSTANDING</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> Natural language interfaces to computers must deal with wide variation in real-word input. Since real-world input is often missing words and contains variant syntax, a useful natural language interface must understand such input. Unfortunately, the technology needed to provide this robustness is not fully mature, and there is uncertainty as to the directions in which such maturity lies.</Paragraph>
    <Paragraph position="1"> Part of this uncertainty centers around the relationship between syntax, semantics, and world knowledge in natural language processing. One theoretical position, the integrated processing hypothesis (Schank 1981), describes an approach to computer modelling of human language processing, based on the idea that these three sorts of knowledge must be applied together and interactively. Since human language understanding is robust, a computer model of human language processing based on this position should also be robust.</Paragraph>
    <Paragraph position="2"> This paper describes a research project designed to explore this conjecture, and describes a robust natural language interface, called MURPHY, which embodies the integrated processing hypothesis. It argues that integrated processing yields robustness and that this hypothesis therefore represents a promising approach to the construction of robust natural language interfaces.</Paragraph>
    <Paragraph position="3"> MURPHY has been developed within a limited domain, and questions remain about the generality of its techniques. Nonetheless, its performance within this domain suggests that generalization to a richer and more realistic domain is possible.</Paragraph>
    <Paragraph position="4"> This paper first defines the term robust as it will be used here, and introduces the MURPHY system. Second, it considers previous work on the problems of robustness.</Paragraph>
    <Paragraph position="5"> Third, it describes the integrated processing hypothesis and the motivation for the research strategy adopted here. Fourth, it describes the MURPHY system and its performance in detail, and then argues that this performance derives from its implementation of the integrated processing hypothesis. Finally, it suggests that the integrated processing hypothesis is indeed a promising approach to the construction of robust natural language interfaces, and that MURPHY represents a successful first step.</Paragraph>
  </Section>
class="xml-element"></Paper>