<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1614">
  <Title>Computational Mechanisms for Pun Generation</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have seen that existing pun generators assume a relatively simple rule-based generation architecture, but the way that this architecture has been used is not typical of NLG tasks.</Paragraph>
    <Paragraph position="1"> These generators produce self-contained puns, but such jokes would be of potential use only in certain applications (where bald joke-telling is acceptable, or where language is to be studied rather than used in natural communication). The case for using contextually integrated puns to enhance computer systems is more plausible, but in the most general case this could lead to computational complexity. Nevertheless, there may be some tractable &amp;quot;tricks&amp;quot; which an NLG system could use to produce very simple contextually-integrated puns.</Paragraph>
    <Paragraph position="2"> Acknowledgements: This paper benefitted from comments from the Aberdeen NLG group, and from Ruli Manurung.</Paragraph>
  </Section>
class="xml-element"></Paper>