<?xml version="1.0" standalone="yes"?>
<Paper uid="P04-1011">
  <Title>Trainable Sentence Planning for Complex Information Presentation in Spoken Dialog Systems</Title>
  <Section position="14" start_page="4" end_page="4" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> This paper shows that the training technique used in SPoT can be easily extended to a new  test fold which have the largest negative impact on the nal RankBoost score (above the double line) and the largest positive impact on the nal RankBoost score (below the double line), for Compare3. s represents the increment or decrement associated with satisfying the condition. null domain and used for information presentation as well as information gathering. Previous work on SPoT also compared trainable sentence planning to a template-based generator that had previously been developed for the same application (Rambow et al., 2001). The evaluation results for SPaRKy (1) support the results for SPoT, by showing that trainable sentence generation can produce output comparable to template-based generation, even for complex information presentations such as extended comparisons; (2) show that trainable sentence generation is sensitive to variations in domain application, presentation type, and even human preferences about the arrangement of particular types of information.</Paragraph>
  </Section>
class="xml-element"></Paper>