<?xml version="1.0" standalone="yes"?>
<Paper uid="H91-1045">
  <Title>Calculating the Probability of a Partial Parse of a Sentence</Title>
  <Section position="7" start_page="239" end_page="240" type="concl">
    <SectionTitle>
CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> With the addition of this score, there are now a number of different methods for controlling the parsing of sentences from a stochastic grammar, each with its own kind of parser and expected form of the grammar. The four we know of are: \[1, 5, 8, 9\]. It is possible to find &amp;quot;expensive&amp;quot; grammars for each of these  scores. For our score, a &amp;quot;cheap&amp;quot; grammar is one in which each symbol is the left child in relatively few rules.</Paragraph>
    <Paragraph position="1"> The goal, then, must be to find a parser, score and grammar that meet the needs of a particular application. We take at least some small comfort from the fact that our score has a Bayesian &amp;quot;maximum likelihood&amp;quot; interpretation, even though the superiority of that approach depends on the shaky assumption that the input being parsed really is the randomly-generated output of the stochastic grammar under consideration.</Paragraph>
  </Section>
class="xml-element"></Paper>