<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1513">
  <Title>Three reasons to adopt TAG-based surface realisation</Title>
  <Section position="8" start_page="101" end_page="101" type="concl">
    <SectionTitle>
7 Further research
</SectionTitle>
    <Paragraph position="0"> Although the results presented give strong evidence for the claim that TAG naturally supports the development of an optimised surface based realiser, they are based on a limited testsuite and on a core grammar for French that heavily overgenerates. Hence they do not truly reflect the potential of the proposed optimisations on the performance of a large scale surface realiser. Current work concentrates on remedying these shortcomings. In particular, we are working on developing a structured test suite which permits a precise measure of the impact of different factors both on complexity and on the optimisations used. In this testsuite for instance, each item is associated with a series of indicators concerning its potential complexity: number of literals in the corresponding input semantics, number of trees, number of nodes, number of substitutions nodes and number of foot nodes in the corresponding selection of initial trees.</Paragraph>
    <Paragraph position="1"> Further work also includes restricting overgeneration and exploring in how far, polarity filtering can be used to select one among the many paraphrases null</Paragraph>
  </Section>
class="xml-element"></Paper>