<?xml version="1.0" standalone="yes"?>
<Paper uid="H05-1031">
  <Title>Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP), pages 241-248, Vancouver, October 2005. c(c)2005 Association for Computational Linguistics Automatically Learning Cognitive Status for Multi-Document Summarization of Newswire</Title>
  <Section position="4" start_page="246" end_page="247" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> Cognitive status distinctions are important when generating summaries, as they help determine both  what to say and how to say it. However, to date, no one has attempted the task of inferring cognitive status from unrestricted news.</Paragraph>
    <Paragraph position="1"> We have shown that the hearer-old/new and major/minor distinctions can be inferred using features derived from the lexical and syntactic forms and frequencies of references in the news reports. We have presented results that show agreement on the familiarity distinction between educated adult American readers with an interest in current affairs, and that the learned classi er accurately predicts this distinction. We have demonstrated that the acquired cognitive status is useful for determining which characters to name in summaries, and which named characters to describe or elaborate. This provides the foundation for a principled framework in which to address the question of how much references can be shortened without compromising readability.</Paragraph>
  </Section>
class="xml-element"></Paper>