<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1513">
  <Title>Three reasons to adopt TAG-based surface realisation</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Surface realisation consists in producing all the sentences associated by a grammar with a given semantic formula. For lexicalist grammars such as LTAG (Lexicalised Tree Adjoining Grammar), surface realisation usually proceeds bottom-up from a set of flat semantic literals1. However, surface realisation from flat semantic formulae is known to be exponential in the length of the input (Kay96; Bre92; KS02). In this paper, we abstract from the TAG based surface realiser for French GenI, (GK05) and argue that TAG naturally supports the integration of various proposals made to help reduce either surface realisation or parsing complexity into a TAG based, lexically driven surface realiser. Specifically, we show: 1. that TAG elementary trees naturally support the implementation of a technique called polarity filtering used to reduce the exponential factor introduced by lexical ambiguity (Per03), 1See e.g., (CCFP99) for a discussion summarising the reasons for this choice.</Paragraph>
    <Paragraph position="1"> 2. that TAG two operations of substitution and adjunction provides a natural framework for implementing a delayed adjunction mechanism capable of reducing the complexity due to the lack of ordering information and 3. that TAG extended domain of locality helps reduce the potential complexity increment introduced by semantically empty items such as infinitival &amp;quot;to&amp;quot; or complementiser &amp;quot;that&amp;quot;.</Paragraph>
  </Section>
class="xml-element"></Paper>