<?xml version="1.0" standalone="yes"?>
<Paper uid="N03-2024">
  <Title>References to Named Entities: a Corpus Study</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion and Future work
</SectionTitle>
    <Paragraph position="0"> As has been seen, a major improvement of summary readability can be achieved by using the simple set of rewrite rules that realize the highest probability path in the derived Markov model. One possible usage of the model which is not discussed in the paper but is the focus of current and ongoing work, is to generate realizations &amp;quot;on demand&amp;quot;. Referring expressions can be generated by recombining different pieces of the input rather than the currently used extraction of full NPs. This approach will make better use of the Markov model, but it also requires work towards deeper semantic processing of the input.</Paragraph>
    <Paragraph position="1"> Semantic information is needed in order to prevent the combination of almost synonymous premodifiers in the same NP and also for the identification of properties that are more central for the enity with respect to the focus of the input cluster.</Paragraph>
  </Section>
class="xml-element"></Paper>