<?xml version="1.0" standalone="yes"?>
<Paper uid="P05-1022">
  <Title>Coarse-to-fine n-best parsing and MaxEnt discriminative reranking</Title>
  <Section position="6" start_page="178" end_page="178" type="evalu">
    <SectionTitle>
5 Experimental results
</SectionTitle>
    <Paragraph position="0"> We evaluated the performance of our reranking parser using the standard PARSEVAL metrics. We n-best trees f-score  best trees, with weights estimated from sections 2-21 and the regularizer constant c adjusted for optimal f-score on section 24 and evaluated on sentences of length less than 100 in section 23. trained the n-best parser on sections 2-21 of the Penn Treebank, and used section 24 as development data to tune the mixing parameters of the smoothing model. Similarly, we trained the feature weights th with the MaxEnt reranker on sections 2-21, and adjusted the regularizer constant c to maximize the f-score on section 24 of the treebank. We did this both on the trees supplied to us by Michael Collins, and on the output of the n-best parser described in this paper. The results are presented in Table 3. The n-best parser's most probable parses are already of state-of-the-art quality, but the reranker further improves the f-score.</Paragraph>
  </Section>
class="xml-element"></Paper>