<?xml version="1.0" standalone="yes"?>
<Paper uid="W01-1203">
  <Title>Parsing and Question Classification for Question Answering</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> We showed that question parsing dramatically improves when complementing the Penn treebank training corpus with an additional treebank of 1153 questions. We described the different answer types (&amp;quot;Qtargets&amp;quot;) that questions are classified as and presented how we semantically enriched parse trees to facilitate question-answer matching.</Paragraph>
    <Paragraph position="1"> Even though we started our Webclopedia project only five months before the TREC9 evaluation, our Q&amp;A system received an overall Mean Reciprocal Rank of 0.318, which put Webclopedia in essentially tied second place with two others. (The best system far outperformed those in second place.) During the TREC9 evaluation, our deterministic (and therefore time-linear) CONTEX parser robustly parsed approximately 250,000 sentences, successfully producing a full parse tree for each one of them.</Paragraph>
    <Paragraph position="2"> Since then we scaled up question treebank from 250 to 1153; roughly doubled the number of Qtarget types and rules; added more features to the machine-learning based parser; did some more treebank cleaning; and added more background knowledge to our ontology.</Paragraph>
    <Paragraph position="3"> In the future, we plan to refine the Qtarget hierarchy even further and hope to acquire Qtarget rules through learning.</Paragraph>
    <Paragraph position="4"> We plan to make the question treebank publicly available.</Paragraph>
  </Section>
class="xml-element"></Paper>