<?xml version="1.0" standalone="yes"?>
<Paper uid="P94-1026">
  <Title>GRAMMAR SPECIALIZATION THROUGH ENTROPY THRESHOLDS</Title>
  <Section position="9" start_page="193" end_page="194" type="concl">
    <SectionTitle>
SUMMARY
</SectionTitle>
    <Paragraph position="0"> This article proposes a method for automatically finding the appropriate tree-cutting criteria in the EBG scheme, rather than having to hand-code them. The EBG scheme has previously proved most successful for  tuning a natural-language grammar to a specific application domain and thereby achieve very much faster parsing, at the cost of a small reduction in coverage. Instruments have been developed and tested for controlling the coverage and for avoiding a large number of short reductions, which is argued to be the main source to poor parser performance. Although these instruments are currently slightly too blunt to enable producing grammars with the same high performance as the hand-coded tree-cutting criteria, they can most probably be sharpened by future research, and in particular refined to achieve the delicate balance between high coverage and a distribution of reduction lengths that is sufficiently biased towards long reductions. Also, banning recursion by category specialization, i.e. by for example distinguishing NPs that dominate other NPs from those that do not, will he investigated, since this is believed to be an important ingredient in the version of the scheme employing hand-coded tree-cutting criteria.</Paragraph>
  </Section>
class="xml-element"></Paper>