<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0603">
  <Title>Search Engine Statistics Beyond the n-gram: Application to Noun Compound Bracketing</Title>
  <Section position="8" start_page="23" end_page="23" type="concl">
    <SectionTitle>
5 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> We have extended and improved upon the state-of-the-art approaches to NC bracketing using an unsupervised method that is more robust than Lauer (1995) and more accurate than Lapata and Keller (2004). Future work will include testing on NCs consisting of more than 3 nouns, recognizing the ambiguous cases, and bracketing NPs that include determiners and modifiers. We plan to test this approach on other important NLP problems.</Paragraph>
    <Paragraph position="1"> As mentioned above, NC bracketing should be helpful for semantic interpretation. Another possible application is the refinement of parser output. Currently, NPs in the Penn TreeBank are flat, without internal structure. Absent any other information, probabilistic parsers typically assume right bracketing, which is incorrect about 2/3rds of the time for 3-noun NCs. It may be useful to augment the Penn TreeBank with dependencies inside the currently flat NPs, which may improve their performance overall.</Paragraph>
  </Section>
class="xml-element"></Paper>