<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1208">
  <Title>A Probabilistic Setting and Lexical Cooccurrence Model for Textual Entailment</Title>
  <Section position="8" start_page="47" end_page="47" type="evalu">
    <SectionTitle>
5.2 Results
</SectionTitle>
    <Paragraph position="0"> precision at top-20 and the average confidence weighted score (cws) achieved for the 50 hypotheses. Applying Wilcoxon Signed-Rank Test, our model performs significantly better (at the 0.01 level) than entscore and base for both precision and cws. Analyzing the results showed that many of the mistakes were not due to wrong expansion but rather to a lack of a deeper analysis of the text and hypothesis (e.g. example 3 in Table 2). Indeed this is a common problem with lexical models. Incorporating additional linguistic levels into the probabilistic entailment model, such as syntactic matching, co-reference resolution and word sense disambiguation, becomes a challenging target for future research.</Paragraph>
  </Section>
class="xml-element"></Paper>