<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-1205">
  <Title>An evolutionary approach for improving the quality of automatic summaries</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper we presented two algorithms which combine content information with context information. The first one is a greedy algorithm which chooses one sentence at a time, but once a sentence is selected it cannot be discarded.</Paragraph>
    <Paragraph position="1"> The second algorithm employs an evolutionary technique to determine the set of extracted sentences, overcoming the limitations of the first algorithm.</Paragraph>
    <Paragraph position="2">  evolutionary method performs consistently better than the rest of the methods in terms of coherence and the cohesion, and does not degrade the information content in most of the cases.</Paragraph>
    <Paragraph position="3"> From each text we produced 3% and 5% summaries. For the 3% summaries there is no significant improvement when contextual information is used (not even when the evolutionary algorithm is used). However, for 5% summaries, the number of discourse ruptures in the summaries produced by the evolutionary algorithm is almost half the number of DR in the ones produced by the basic method. The number of dangling referential expressions also reduces. Regardless the length of the summary, it seems to be no significant difference between the basic method and the greedy algorithm. One could argue that for long documents, 5% summaries are too long, and that shorter versions are required. This is true, but these summaries can be shortened by using aggregation rules like the ones proposed in (Otterbacher et al., 2002), where two sentences referring to the same entity are merged into one. Given that the summaries produced with the evolutionary algorithm contain more sequences of sentences related to the same entity, it will be easier to apply such aggregation rules.</Paragraph>
    <Paragraph position="4"> As noted in Section 5.1, the results vary from one text to another. In some cases the continuity principle noticeably improves the quality of a summary, but in other cases the improvement is moderate or low. One reason could be that the continuity principle alone is too weak to be able to guarantee the coherence of the produced summary. We intend to extend our experiments and test whether a combination of centering theory's principles, as used in (Kibble and Power, 2000), can lead to better results.</Paragraph>
    <Paragraph position="5"> Our algorithms were tested on scientific articles. We intend to extend the evaluation using other types of texts in order to learn if the genre influences the results.</Paragraph>
    <Paragraph position="6"> To conclude, in this paper we argue that it is possible to improve the quality of automatic summaries by using the continuity principle and by employing an evolutionary algorithm to select sentences. This improvement seems to be text dependent, in some cases being small.</Paragraph>
  </Section>
class="xml-element"></Paper>