<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-1124">
  <Title>Improving summarization through rhetorical parsing tuning</Title>
  <Section position="6" start_page="214" end_page="214" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> The empirical and computational experiments that we described in this paper suppon at least the following conclusions. 1. For extracting 10% summaries of short articles of the news story genre, a simple lead-based algorithm is the most efficient solution. 2. For extracting longer summaries of short newspaper articles and for extracting any size summaries of complex (not necessarily news stories) newspaper articles, a simple lead-based algorithm does not provide a satisfactory solution. This assertion holds for other text genres, such as that of Scientific American, as well. 3. There is no magic key (heuristic) for obraining good summarization results; rather the strength of a summarization system seems to come from its ability to combine a multitude of heuristics. 4. Combinations of heuristics that yield &amp;quot;optimal&amp;quot; results for certain summary extract lengths might not yield optimal results for different lengths. 5. Incorporating various heuristics into a discourse-based summarization framework yields good results. 6. In order to assess confidently the effectiveness of the summarization methodology that was introduced here, much larger corpora are required.</Paragraph>
    <Paragraph position="1"> Acknowledgements. I am grateful to Hongyan Jing, Regina Barzilay, Kathleen McKeown, and Michael E1hadad for sharing their TREC-based summarization corpus and to Chin-Yew Lin for fruitful discussions of previous versions of this paper. Special thanks go to Eduard Hovy for comments, discussions, and stimulative enthusiasm. null</Paragraph>
  </Section>
class="xml-element"></Paper>