<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-1006">
  <Title>Learning to recognize features of valid textual entailments</Title>
  <Section position="9" start_page="46" end_page="47" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> The best current approaches to the problem of textual inference work by aligning semantic graphs,  favor entailment. Weights near 0 are omitted. Based on training on the PASCAL RTE development set.</Paragraph>
    <Paragraph position="1"> using a locally-decomposable alignment score as a proxy for strength of entailment. We have argued that such models suffer from three crucial limitations: an assumption of monotonicity, an assumption of locality, and a confounding of alignment and entailment determination.</Paragraph>
    <Paragraph position="2"> We have described a system which extends alignment-based systems while attempting to address these limitations. After finding the best alignment between text and hypothesis, we extract high-level semantic features of the entailment problem, and input these features to a statistical classifier to make an entailment decision. Using this multi-stage architecture, we report results on the PASCAL RTE data which surpass previously-reported results for alignment-based systems.</Paragraph>
    <Paragraph position="3"> We see the present work as a first step in a promising direction. Much work remains in improving the entailment features, many of which may be seen as rough approximations to a formal monotonicity calculus. In future, we aim to combine more precise modeling of monotonicity effects with better modeling of paraphrase equivalence.</Paragraph>
  </Section>
class="xml-element"></Paper>