<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-2902">
  <Title>Porting Statistical Parsers with Data-Defined Kernels</Title>
  <Section position="9" start_page="12" end_page="12" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> This paper proposes a novel technique for improving parser portability, applying parse reranking with data-defined kernels. First a probabilistic model of parsing is trained on all the available data, including a large set of data from the source domain. This model is used to define a kernel over parse trees.</Paragraph>
    <Paragraph position="1"> Then this kernel is used in a large margin classifier 7The sizes of Brown sections reported in (Ratnaparkhi, 1999) do not match the sizes of sections distributed in the Penn Treebank 3.0 package, so we couldn't replicate their split. We suspect that a preliminary version of the corpus was used for their experiments.</Paragraph>
    <Paragraph position="2"> trained on a small set of data only from the target domain. This classifier is used to rerank the top parses produced by the probabilistic model on the target domain. Experiments with a neural network statistical parser demonstrate that this approach leads to improved parser accuracy on the target domain, without any significant increase in computational cost.</Paragraph>
  </Section>
class="xml-element"></Paper>