<?xml version="1.0" standalone="yes"?>
<Paper uid="P84-1089">
  <Title>Coping with Extragrarnmaticality</Title>
  <Section position="7" start_page="441" end_page="442" type="concl">
    <SectionTitle>
6. Concluding Remarks
</SectionTitle>
    <Paragraph position="0"> Any practical natural language interface must be capable of dealing with a wide range of extragrammatical input. This paper has proposed a partial taxonomy of extragrammatica!!ties that arise in spontaneously generated input to a restricted-domain natural language interface and has presented recovery strategies for handhng many of the categories. We also discussed how well three widely employed approaches to parsing -- network-based parsing, pattern matching, and case frame instantation -- could support the recovery strategies, and concluded that case frame instantiation provided the best basis The reader is referred to \[8\]  for a more complete presentation, including a more complete taxonomy and additional recovery strategies, particularly at the dialogue level.</Paragraph>
    <Paragraph position="1"> Based on the set of recovery strategies we have examined and the problems that arise in trying to integrate them with techniques for parsing grammatical input, we offer the following set of desiderata for a parsing process that has to deal with extragrammatical input: = The parsing process should be as interpretive as possible. We have seen several times the need for a parsing process to &amp;quot;stand back&amp;quot; and look at the broad picture of the set of expectations (or grammar) it is applying to the input when an ungrammaticality arises. The more interpretive a parser is, tbe better able it is to do this. A highly interpretive parser is also better able to apply its expectations to the input in more than one way, which may be crucial if the standard way does not work in the face of an ungrammaticality.</Paragraph>
    <Paragraph position="2"> * The parsing process should make it easy to apply semantic information. As we have seen, semantic information is often very important in resolving ungrammaticalities.</Paragraph>
    <Paragraph position="3"> = The parsing process should be able to take advantage of non-uniformity in language like that identified in Section 4.2. As we have seen, recovery can be much more efficient and reliable if a parser is able to make use of variations in ease of recognition or discriminating power between different constituents. Th~s kind of &amp;quot;opportunism&amp;quot; can be built into recovery strategies.</Paragraph>
    <Paragraph position="4"> = The parsing process should be capable of operating top.</Paragraph>
    <Paragraph position="5"> down as well as bottom-up. We have seen examples where both of these modes are essential.</Paragraph>
    <Paragraph position="6"> We believe that case frame mstantiation provides a better basis for parsing extragrammatical input than network-based parsing or pat!ern matching precisely because it satisfies these desiderata better than the other two approaches. We also believe that it is possible do even better than case frame instantiation by using a multi-strategy approach in which case frame instantiation is just one member (albeit a very important one) of a whole array of parsiag and recovery strategies. We argue this claim in detail in \[8,\] and support it by discussion of three experimental parsers that in varying degrees adopt the multi-strategy approach.</Paragraph>
  </Section>
class="xml-element"></Paper>