<?xml version="1.0" standalone="yes"?>
<Paper uid="P05-1011">
  <Title>Probabilistic disambiguation models for wide-coverage HPSG parsing</Title>
  <Section position="18" start_page="88" end_page="89" type="concl">
    <SectionTitle>
7 Discussion and related work
</SectionTitle>
    <Paragraph position="0"> Experiments on deep parsing of Penn Treebank have been reported for Combinatory Categorial Grammar (CCG) (Clark and Curran, 2004b) and Lexical Functional Grammar (LFG) (Kaplan et al., 2004). They developed log-linear models on a packed representation of parse forests, which is similar to our representation. Although HPSG exploits further complicated feature constraints and requires high com- null putational cost, our work has proved that log-linear models can be applied to HPSG parsing and attain accurate and wide-coverage parsing.</Paragraph>
    <Paragraph position="1"> Clark and Curran (2004a) described a method of reducing the cost of parsing a training treebank in the context of CCG parsing. They first assigned to each word a small number of supertags, which correspond to lexical entries in our case, and parsed supertagged sentences. Since they did not mention the probabilities of supertags, their method corresponds to our &amp;quot;filtering only&amp;quot; method. However, they also applied the same supertagger in a parsing stage, and this seemed to be crucial for high accuracy. This means that they estimated the probability of producing a parse tree from a supertagged sentence.</Paragraph>
    <Paragraph position="2"> Another approach to estimating log-linear models for HPSG is to extract a small informative sample from the original set CCB4D7B5 (Osborne, 2000).</Paragraph>
    <Paragraph position="3"> Malouf and van Noord (2004) successfully applied this method to German HPSG. The problem with this method was in the approximation of exponentially many parse trees by a polynomial-size sample. However, their method has the advantage that any features on a parse tree can be incorporated into the model. The trade-off between approximation and locality of features is an outstanding problem.</Paragraph>
    <Paragraph position="4"> Other discriminative classifiers were applied to the disambiguation in HPSG parsing (Baldridge and Osborne, 2003; Toutanova et al., 2004). The problem of exponential explosion is also inevitable for  their methods. An approach similar to ours may be applied to them, following the study on the learning of a discriminative classifier for a packed representation (Taskar et al., 2004).</Paragraph>
    <Paragraph position="5"> As discussed in Section 6, exploration of other features is indispensable to further improvements.</Paragraph>
    <Paragraph position="6"> A possible direction is to encode larger contexts of parse trees, which were shown to improve the accuracy (Toutanova and Manning, 2002; Toutanova et al., 2004). Future work includes the investigation of such features, as well as the abstraction of lexical dependencies like semantic classes.</Paragraph>
  </Section>
class="xml-element"></Paper>