<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1620">
  <Title>An Evolutionary Approach to Referring Expression Generation and Aggregation</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions and future work
</SectionTitle>
    <Paragraph position="0"> With respect to both of the tasks addressed, the output texts respect the specific constraints required for the text to be acceptable, while at the same time showing reasonable variation between the different options much as a human-generated text would. We are working on extending the system to allow the use of proper nouns to describe some concepts, as an additional option to pronouns and descriptive references, including the revision of the genetic operators and the introduction of new evaluation functions to estimate the correct application of proper nouns.</Paragraph>
    <Paragraph position="1"> In view of these results, in future work we want to apply EA techniques to other tasks of NLG, such as content determination and discourse planning. The particular advantages of evolutionary techniques, combined stage by stage in this manner, may be an extremely powerful method for solving natural language generation problems while also profiting from classic NLG techniques.</Paragraph>
    <Paragraph position="2"> It would be also interesting to compare our solution with different approaches found in the literature, as for example [Reiter and Dale, 1992] or [Krahmer and Theune, 2000] for the referring expression generation, and the one of Dalianis and Hovy [Dalianis and Hovy, 1996] for the aggregation.</Paragraph>
    <Paragraph position="3"> Finally, an evaluation as the one proposed in [Callaway and Lester, 2001] would be useful to estimate the goodness of the generated texts. The authors describe the evaluation of STO-RYBOOK, a narrative prose generation system that produces original fairy tales in the Little Red Riding Hood domain.</Paragraph>
    <Paragraph position="4"> They pretend to evaluate multiple versions of a single story assuring that the content is identical across them. Five versions of two separate stories are produced, a pool of twenty students in English compare them, and at last they are analyzed with an ANOVA test.</Paragraph>
  </Section>
class="xml-element"></Paper>