<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0636">
  <Title>Joint Parsing and Semantic Role Labeling</Title>
  <Section position="7" start_page="227" end_page="227" type="concl">
    <SectionTitle>
6 Conclusion and Related Work
</SectionTitle>
    <Paragraph position="0"> In this paper, we have considered several methods for reranking parse trees using information from semantic role labeling. So far, we have not been able to show improvement over selecting the 1-best parse tree. Gildea and Jurafsky (Gildea and Jurafsky, 2002) also report results on reranking parses using an SRL system, with negative results. In this paper, we confirm these results with a MaxEnt-trained SRL model, and we extend them to show that weighting the probabilities does not help either.</Paragraph>
    <Paragraph position="1"> Our results with Collins-style reranking are too preliminary to draw definite conclusions, but the potential improvement does not appear to be great. In future work, we will explore the max-sum approach, which has promise to avoid the pitfalls of max-max reranking approaches.</Paragraph>
  </Section>
class="xml-element"></Paper>