<?xml version="1.0" standalone="yes"?>
<Paper uid="P05-1027">
  <Title>Question Answering as Question-Biased Term Extraction: A New Approach toward Multilingual QA</Title>
  <Section position="3" start_page="215" end_page="216" type="intro">
    <SectionTitle>
2 Preparation
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="215" end_page="215" type="sub_section">
      <SectionTitle>
2.1 Training Data
</SectionTitle>
      <Paragraph position="0"> prises 2,000 Japanese questions with correct answers as well as question types and IDs of articles that contain the answers. Each question is categorized as one of 115 hierarchically classi ed question types.</Paragraph>
      <Paragraph position="1"> The document set is used not only in the training phase but also in the execution phrase.</Paragraph>
      <Paragraph position="2"> Although the CRL QA Data contains question types, the information of question types are not used for the training. This is because more than the 60% of question types have fewer than 10 questions as examples (Table 1). This means it is very unlikely that we can train a QA system that can handle this 60% due to data sparseness. 2 Only for the purpose of analyzing experimental results in this paper do we refer to the question types of the dataset.</Paragraph>
    </Section>
    <Section position="2" start_page="215" end_page="216" type="sub_section">
      <SectionTitle>
2.2 Learning with Maximum Entropy Models
</SectionTitle>
      <Paragraph position="0"> This section brie y introduces the machine learning technique Maximum Entropy Models and describes how to apply MEMs to QA tasks.</Paragraph>
      <Paragraph position="1">  Let X be a set of input symbols and Y be a set of class labels. A sample (x, y) is a pair of input x={x1,. . . , xm} (xi [?] X) and output y [?] Y.  ysis was reported in (Suzuki et al., 2003), but training and maintaining an answer extractor for question types of ne granularity is not an easy task.</Paragraph>
      <Paragraph position="2">  The Maximum Entropy Principle (Berger et al., 1996) is to nd a model p[?] = argmax p[?]C H(p), which means a probability model p(y|x) that maximizes entropy H(p).</Paragraph>
      <Paragraph position="3"> Given data (x(1), y(1)),. . .,(x(n), y(n)), letuniondisplay</Paragraph>
      <Paragraph position="5"> &lt;~xm, ~ym&gt; }. This means that we enumerate all pairs of an input symbol and label and represent them as &lt;~xi, ~yi&gt; using index i (1 [?] i [?] m).</Paragraph>
      <Paragraph position="6"> In this paper, feature function fi is de ned as follows. null</Paragraph>
      <Paragraph position="8"> We use all combinations of input symbols in x and class labels for features (or the feature function) of MEMs.</Paragraph>
      <Paragraph position="9"> With Lagrangian l = l1, ..., lm, the dual function of H is:</Paragraph>
      <Paragraph position="11"> and ~p(fi) indicate the empirical distribution of x and fi in the training data.</Paragraph>
      <Paragraph position="12"> The dual optimization problem l[?] = argmax l Ps(l) can be ef ciently solved as an optimization problem without constraints. As a result, probabilistic model p[?] = pl[?] is obtained as:</Paragraph>
      <Paragraph position="14"> Question analysis is a classi cation problem that classi es questions into different question types.</Paragraph>
      <Paragraph position="15"> Answer candidate extraction is also a classi cation problem that classi es words into answer types (i.e., question types), such as PERSON, DATE, and AWARD. Answer selection is an exactly classi cation that classi es answer candidates as positive or negative. Therefore, we can apply machine learning techniques to generate classi ers that work as components of a QA system.</Paragraph>
      <Paragraph position="16"> In the QBTE approach, these three components, i.e., question analysis, answer candidate extraction, and answer selection, are integrated into one classier. null To successfully carry out this goal, we have to extract features that re ect properties of correct answers of a question in the context of articles.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>