<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-2030">
  <Title>Feature Selection for a Rich HPSG Grammar Using Decision Trees</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> We have presented work on building probabilistic models for HPSG parse disambiguation. As the number of available features increases it becomes more important to select relevant features automatically. We have shown that decision trees using different feature subspaces can be combined in ensembles that choose the correct parse with higher accuracy than individual models.</Paragraph>
    <Paragraph position="1"> Our plans for future work include exploring more information from the HPSG signs and defining features that capture long distance dependencies. Another line of future research is defining models over the deeper MRS semantic representations, possibly in conjunction with clustering of semantic types.</Paragraph>
  </Section>
class="xml-element"></Paper>