<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-1314">
  <Title>guistic Engineering Environment using LFG (Lexical Functional Grammar) and CG</Title>
  <Section position="7" start_page="97" end_page="98" type="evalu">
    <SectionTitle>
6 Results
</SectionTitle>
    <Paragraph position="0"> The system is implemented in Smalltalk, and its user-friendly interface permits step-by-step monitoring of the process as well as parameter tuning.</Paragraph>
    <Paragraph position="1"> Current work concerns Stendhal's text, as it has the highest density and variety of pronouns (cf. SS 4.1). Its syntactic complexity and the overproductivity of the local LFG parser oblige us to make manual selection among NPs, and disambiguation of the /le/, /In/, /1'/ pronouns vs. definite articles. The next table summarizes our first experiment.</Paragraph>
    <Paragraph position="2">  When coreference is not dealt with, there are as many &amp;quot;characters&amp;quot; as NPs, and, as expected, the number of correctly solved pronouns is smaller (by 40%).</Paragraph>
    <Paragraph position="3"> The pronoun resolution score, 62%, is a little smaller than those obtained elsewhere for English texts; but these results are encouraging, especially as they rely only on simple rules. Moreover, on this particular text, we have observed, that 50% of the mistakes could be avoided using on the one hand simple semantic constraints derived from verbal argument structure (e.g., human/non-human subject, animated/non-animated subject/object...); on the other hand syntactic constraints concerning the possibilities of coreference between NPs and pronouns occurring in the same sentence.</Paragraph>
    <Paragraph position="4"> Further work will first concem a more accurate tuning of the parameters, and adjunction of new activation and selection rules. In particular, syntactic restrictions will be adapted to the local parser's data. Also, we would like to make the processing entirely automatic, which requires a selection among the NPs provided by the lOcal parser, and disambiguation of the pronouns. These being complex tasks, they will probably decrease the success rate, especially with respect to English, where articles and pronouns are never homonymous.</Paragraph>
  </Section>
class="xml-element"></Paper>