<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1661">
  <Title>Statistical Ranking in Tactical Generation</Title>
  <Section position="4" start_page="0" end_page="517" type="intro">
    <SectionTitle>
2 Models
</SectionTitle>
    <Paragraph position="0"> In this section we briefly review the different types of statistical models that we use for ranking the output of the generator. We start by describing the language model, and then go on to review the framework for discriminative MaxEnt models and SVM rankers. In the following we will use D7 and D6 to denote semantic inputs and generated realizations respectively.</Paragraph>
    <Paragraph position="1">  Remember that dogs must be on a leash.</Paragraph>
    <Paragraph position="2"> Remember dogs must be on a leash.</Paragraph>
    <Paragraph position="3"> On a leash, remember that dogs must be.</Paragraph>
    <Paragraph position="4"> On a leash, remember dogs must be.</Paragraph>
    <Paragraph position="5"> A leash, remember that dogs must be on.</Paragraph>
    <Paragraph position="6"> A leash, remember dogs must be on.</Paragraph>
    <Paragraph position="7"> Dogs, remember must be on a leash.</Paragraph>
    <Paragraph position="8">  puts using the ERG. Where the input semantics is no specified for aspects of information structure (e.g. requesting foregrounding of a specific entity), paraphrases include all grammatically legitimate topicalizations. Other choices involve, for example, the optionality of complementizers and relative pronouns, permutation of (intersective) modifiers, and lexical and orthographic alternations.</Paragraph>
    <Section position="1" start_page="517" end_page="517" type="sub_section">
      <SectionTitle>
2.1 Language Models
</SectionTitle>
      <Paragraph position="0"> The use of D2-gram language models is the most common approach to statistical selection in generation (Langkilde &amp; Knight, 1998; and White (2004); inter alios). In order to better assert the relative performance of the discriminative models and the structural features we present below, we also apply a trigram model to the ranking problem. Using the freely available CMU SLM Toolkit (Clarkson &amp; Rosenfeld, 1997), we trained a trigram model on an unannotated version of the British National Corpus (BNC), containing roughly 100 million words (using Witten-Bell discounting and back-off). Given such a model D4  Although in this case scoring is not conditioned on the input semantics at all, we still include it to make the function formulation more general as we will be reusing it later.</Paragraph>
      <Paragraph position="1"> Note that, as the realizations in our symmetric treebank also include punctuation marks, these are also treated as separate tokens by the language model (in addition to pseudo-tokens marking sentence boundaries).</Paragraph>
    </Section>
    <Section position="2" start_page="517" end_page="517" type="sub_section">
      <SectionTitle>
2.2 Maximum Entropy Models
</SectionTitle>
      <Paragraph position="0"> Maximum entropy modeling provides a very flexible framework that has been widely used for a range of tasks in NLP, including parse selection (e.g. Johnson, Geman, Canon, Chi, &amp; Riezler, 1999; Malouf &amp; Noord, 2004) and reranking for machine translation (e.g. Och et al., 2004). A model is specified by a set of real-valued feature functions that describe properties of the data, and an associated set of learned weights that determine the contribution of each feature.</Paragraph>
      <Paragraph position="1"> Let us first introduce some notation before we go on. Let CHB4D7</Paragraph>
      <Paragraph position="3"> CV be the set of realizations licensed by the grammar for a semantic representation D7</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>