<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1023">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Trace Prediction and Recovery With Unlexicalized PCFGs and Slash Features</Title>
  <Section position="7" start_page="182" end_page="183" type="evalu">
    <SectionTitle>
5 Comparison
</SectionTitle>
    <Paragraph position="0"> results of our parser with those reported in Johnson (2001), Dienes and Dubey (2003b) and Campbell (2004). In terms of recall and f-score, our parser outperforms the other parsers. In terms of precision, the tagger of Dienes and Dubey is the best, but its recall is the lowest of all systems.</Paragraph>
    <Paragraph position="1"> prec. recall f-score  on section 23 The good performance of our parser on the empty element recognition task is remarkable considering the fact that its performance on the labeled bracketing task is 3% lower than that of the Charniak (2000) parser used by Campbell (2004). prec. recall f-score this paper 81.7 73.5 77.4  those reported in Johnson (2001), Dienes and Dubey (2003b), Dienes and Dubey (2003a), and Campbell (2004). Our parser achieves the highest precision and f-score. Campbell (2004) reports a higher recall, but lower precision.</Paragraph>
    <Paragraph position="2"> Table 7 shows the trace prediction accuracies of our parser, Johnson's (2001) parser with parser input and perfect input, and Campbell's (2004) parser with perfect input. The accuracy of Johnson's parser is consistently lower than that of the other parsers and it has particular difficulties with ADVP traces, SBAR traces, and empty relative pronouns (WHNP 0). Campbell's parser and our parser cannot be directly compared, but when we take the respective performance difference to Johnson's parser as evidence, we might conclude that Campbell's parser works particularly well on NP *, *U*, and WHNP 0, whereas our system  diction accuracies for different categories in this paper (paper), in (Johnson, 2001) with parser input (J1), in (Johnson, 2001) with perfect input (J2), and in (Campbell, 2004) with perfect input.</Paragraph>
    <Paragraph position="3"> is slightly better on empty complementizers (0), ADVP traces, and SBAR traces.</Paragraph>
  </Section>
class="xml-element"></Paper>