<?xml version="1.0" standalone="yes"?>
<Paper uid="W01-1203">
  <Title>Parsing and Question Classification for Question Answering</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> There has recently been a strong increase in the research of question answering, which identifies and extracts answers from a large collection of text. Unlike information retrieval systems, which return whole documents or larger sections thereof, question answering systems are designed to deliver much more focused answers, e.g.</Paragraph>
    <Paragraph position="1"> Q: Where is Ayer's Rock? A: in central Australia Q: Who was Gennady Lyachin? A: captain of the Russian nuclear submarine Kursk The August 2000 TREC-9 short form Q&amp;A track evaluations, for example, specifically limited answers to 50 bytes.</Paragraph>
    <Paragraph position="2"> The Webclopedia project at the USC Information Sciences Institute (Hovy 2000, 2001) pursues a semantics-based approach to answer pinpointing that relies heavily on parsing. Parsing covers both questions as well as numerous answer sentence candidates. After parsing, exact answers are extracted by matching the parse trees of answer sentence candidates against that of the parsed question. This paper describes the critical challenges that a parser faces in Q&amp;A applications and reports on a number of extensions of a deterministic machine-learning based shift-reduce parser, CONTEX (Hermjakob 1997, 2000), which was previously developed for machine translation applications. In particular, section 2 describes how additional treebanking vastly improved parsing accuracy for questions; section 3 describes how the parse tree is extended to include the answer type of a question, a most critical task in question answering; section 4 presents experimental results for question parsing and QA typing; and finally, section 5 describes how the parse trees of potential answer sentences are enhanced semantically for better question-answer matching.</Paragraph>
  </Section>
class="xml-element"></Paper>