<?xml version="1.0" standalone="yes"?>
<Paper uid="J86-4002">
  <Title>REFERENCE IDENTIFICATION AND REFERENCE IDENTIFICATION FAILURES</Title>
  <Section position="28" start_page="297" end_page="297" type="concl">
    <SectionTitle>
5 CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> This paper had four objectives: * to illustrate how complex reference really is; * to show that previous referent identification paradigms don't suffice, given real world data; * to isolate numerous kinds of knowledge people use for reference resolution; and * to augment current reference algorithms to handle more real descriptions.</Paragraph>
    <Paragraph position="1"> In this section, we provide a summary of our findings and present some reasonable future directions of this work.</Paragraph>
    <Section position="1" start_page="297" end_page="297" type="sub_section">
      <SectionTitle>
5.1 SUMMARY
</SectionTitle>
      <Paragraph position="0"> Our goal in this work is to build robust natural language understanding systems, allowing them to detect and avoid miscommunication. The goal is not to make a perfect listener but to make a more tolerant one that could avoid many mistakes, though it may still be wrong on occasion.</Paragraph>
      <Paragraph position="1"> In this paper, we introduced a taxonomy of miscommunication problems that occur in expert-apprentice dialogues. We showed that reference mistakes are one kind of obstacle to robust communication. To tackle reference errors, we described how to extend the succeed/fail paradigm followed by previous natural language researchers.</Paragraph>
      <Paragraph position="2"> We represented real world objects hierarchically in a knowledge base using a representation language, KL-One, that follows in the tradition of semantic networks and frames. &amp;quot;In such a representation framework, the reference identification task looks for a referent by comparing the representation of the speaker's input to elements in the knowledge base by using a matching procedure. Failure to find a referent in previous reference identification systems resulted in the unsuccessful termination of the reference task. We claim that people behave better than this and explicitly illustrated</Paragraph>
    </Section>
  </Section>
  <Section position="30" start_page="297" end_page="297" type="concl">
    <SectionTitle>
5.2. FUTURE DIRECTIONS
</SectionTitle>
    <Paragraph position="0"> This paper mentioned only a small aspect of what needs to be done with miscommunication. There are much broader problems that we also want to address. We alluded in the paper to problems due to metonomy - the use of the name of one thing for that of another - but never r~ally tried in this work to handle more than a few special cases of it. Consider the three descriptions below. 23 Notice how the noun phrase the window refers to three different things in each utterance.</Paragraph>
    <Paragraph position="1"> The window was broken. (the glass) The window was boarded up. (the opening) Open the window. (the glass/frame inset) Any reasonable reference mechanism must be able to distinguish such differences.</Paragraph>
    <Paragraph position="2"> We neglected to discuss the effect of quantifiers in a description and how they affect the relaxation mechanism. A numerical quantifier could be specified that is incorrect; for example, two pegs when only one is found. Vague quantifiers such as a bit, a few, or a piece can be part of a description. Measurements like a liter of beer or a pound of chicken could also be used. A listener can estimate measurements when looking for the referent or be very precise and measure them. In all these examples, the relaxation of the quantifier could be required before a referent can be found. Our relaxation mechanism could be extended to handle many of these examples.</Paragraph>
    <Paragraph position="3"> The FWIM reference identification system we developed models the reference process by the classification operation of KL-One. We need a more complicated model for reference. That model might need a complete identification plan that requires making inferences beyond those provided by classification. The model could also require the execution of a physical action by the listener before determining the proper referent. Cohen (1984:101) gives two excellent examples of such reference plans. The first, &amp;quot;the magnetic screwdriver, please,&amp;quot; requires the listener to place various screwdrivers against metal to determine which is magnetic. The second, &amp;quot;the three two-inch long salted green noodles&amp;quot; requires the listener to count, examine, measure and taste to discover the proper referent.</Paragraph>
    <Paragraph position="4"> The FWlM reference system uses relaxation rules to compile knowledge source information. These rules provide a convenient forum for evaluating a description with respect to language and physical knowledge about the world. However, reasoning mechanisms that &amp;quot;think&amp;quot; about these knowledge sources sh~ould really replace the rules and become part of the negotiation process.</Paragraph>
    <Paragraph position="5"> There are also miscommunication problems that are outside of the reference area. We need to consider full utterances and the associated discourse in which they appear. Utterances can be imprecise or ill-formed with respect to the current discourse. The goals specified by a speaker through a particular utterance or discourse could be confused. For example, a speaker's requested goal could be outside the scope of the domain being discussed.</Paragraph>
    <Paragraph position="6"> We believe that our model will help solve the problem for this bigger picture. In particular, we feel the negotiation method will be important here, too. The negotiation process will become part of the plan recognition section of a natural language system. There a search of the plan space for the set of plans that might fit the utterance or sequence of utterances would be performed. A relaxation component related in style to the one outlined in this paper could be invoked to provide an orderly relaxation of the speaker's utterances to fit the plans and the Computational Linguistics, Volume 12, Number 4, October-December 1986 303 Bradley A. Goodman Reference Identification and Reference Identification Failures domain world. This process will require more interaction with the speaker through the use of clarification dialogues.</Paragraph>
  </Section>
class="xml-element"></Paper>