<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-2009">
  <Title>Answering the Question You Wish They Had Asked: The Impact of Paraphrasing for Question Answering</Title>
  <Section position="5" start_page="35" end_page="35" type="relat">
    <SectionTitle>
5 Related Work
</SectionTitle>
    <Paragraph position="0"> Most of the work in QA and paraphrasing focused on folding paraphrasing knowledge into the question analyzer or the answer locator (Rinaldi et al., 2003; Tomuro, 2003). Our work, on the contrary, focuses on question paraphrasing as an external component, independent of the QA system architecture.</Paragraph>
    <Paragraph position="1"> Some authors (Dumais et al., 2002; Echihabi et al., 2004) considered the query sent to a search engine as a &amp;quot;paraphrase&amp;quot; of the original natural language question. For instance, Echihabi et al. (2004) presented a large number of &amp;quot;reformulations&amp;quot; that transformed the query into assertions that could match the answers in text. Here we understand a question paraphrase as a reformulation that is itself a question, not a search engine query.</Paragraph>
    <Paragraph position="2"> Other efforts in using paraphrasing for QA (Duclaye et al., 2003) focused on using the Web to obtain different verbalizations for a seed relation (e.g., Author/Book); however, they have yet to apply their learned paraphrases to QA.</Paragraph>
    <Paragraph position="3"> Recently, there has been work on identifying paraphrases equivalence classes for log analysis (Hedstrom, 2005). Hedstrom used a vector model from Information Retrieval that inspired our cosine measure feature described in Section 3.</Paragraph>
  </Section>
class="xml-element"></Paper>