<?xml version="1.0" standalone="yes"?>
<Paper uid="P81-1025">
  <Title>PERSPECTIVES ON PARSING ISSUES</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
COMPUTATIONAL PEESPECTT VE
IS IT USEFUL TO DISTINGUISH PARSING FROM INTERPRETATION?
</SectionTitle>
    <Paragraph position="0"> Since most of this position paper viii be attacking the separation of parsing from interpretation, let me first make it clear that I do believe in syntactic knowledge. In this I am more conservative than other researchers in interpretation at Berkeley, Carnegie-Mellon, Colombia, the universities of Connecticut and Maryland, and Yale.</Paragraph>
    <Paragraph position="1"> But believing in syntactic knowledge is not the same as believing in parsers! The search for a way to assign 8 syntactic structure to a sentence largely independent of the meaning of that sentence has led to a terrible misdirection of labor. And this effect has been felt on both sides of the fence. We find ourselves looking for ways to reduce interaction between syntax and semantics as much as possible. How far can we drive a purely syntactic (semantic) analyzer, without sneaking over into the enemy camp? Row well can we disguise syntax (semantics) as semantics (syntax)? How narrow a pipe between the two can we set away with? What a waste of time, when we should be starting with bodies of texts, considering the total language analysis picture, and looking for what kinds of knowledge need to interact to understand those texts.</Paragraph>
    <Paragraph position="2"> If our intent in overextendins our theories was to rest their muscle, then I would have no qualms. Pushing a mechanism down a blind alley is an important way to study its weaknesses. But I really can't accept this Popperian view of modern computational linguistics.</Paragraph>
    <Paragraph position="3"> Mechanisms are not driven beyond their limits to find those limits, but rather to grab territory from the other side. The underlying premise is &amp;quot;If our mechanism X can sometimes do task A, then there is no need for someone else's mechanism Y.&amp;quot; Occam's razor is used with murderous intent.</Paragraph>
    <Paragraph position="4"> Furthermore, the debate over whether parsers make sense has drastically reduced interaction between researchers. Each side sees the other as avoiding fundamental issues, and so the results from the other side always seem to be beside the point. For example, when Mirth Marcus&amp;quot; explains some grmamatical constraint as syntactic processing constraints, be doesn't answer any of the problems I'm faced with. And I'm sure Mitch has no need for frame-based, domain-driven partial language analysis techniques.</Paragraph>
    <Paragraph position="5"> This situation has not arisen because we have been forced to specialize. We simply don't know enough to qualify for an information explosion yet. Computational linguistics doesn't have hundreds of journals in dozens of languages. It's a young field with only a handful of people working in it.</Paragraph>
    <Paragraph position="6"> Nor is it the case that we don't have things to say to each other. But -- end here's the rub -- some of the most useful things that each of us knows are the things that we don't dare tell. By that I mean that each of us knows where our theories fall apart, where ve have to kludge the programs, fudge the inputs, or wince at the outputs. That kind of information could be invaluable for suggesting to the others where to focus their attentions. Unfortunately, even if we became brave enough to talk about, even emphasize, where we're having problems, the odds are low that we would consider acceptable what someone else proposes as a solution.</Paragraph>
  </Section>
class="xml-element"></Paper>