<?xml version="1.0" standalone="yes"?>
<Paper uid="P01-1010">
  <Title>What is the Minimal Set of Fragments that Achieves Maximal Parse Accuracy?</Title>
  <Section position="6" start_page="1" end_page="2" type="evalu">
    <SectionTitle>
4.6 Results for all sentences
</SectionTitle>
    <Paragraph position="0"> We have seen that for test sentences [?] 40 words, maximal parse accuracy was obtained by a subtree set which is restricted to subtrees with not more than 12 words and which does not contain unlexicalized subtrees deeper than 6.</Paragraph>
    <Paragraph position="1">  We used  It may be noteworthy that for the development set (section 22 of WSJ), maximal parse accuracy was obtained with exactly the same subtree restrictions. As explained in 4.1, we initially tested all restrictions on the development set, but we preferred to report the effects of these restrictions for the test set.</Paragraph>
    <Paragraph position="2"> these restrictions to test our model on all sentences [?] 100 words from the WSJ test set. This resulted in an LP of 89.7% and an LR of 89.7%. These scores slightly outperform the best previously published parser by Charniak (2000), who obtained 89.5% LP and 89.6% LR for test sentences [?] 100 words. Only the reranking technique proposed by Collins (2000) slightly outperforms our precision score, but not our recall score: 89.9% LP and 89.6% LR.</Paragraph>
  </Section>
class="xml-element"></Paper>