<?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="6" start_page="35" end_page="35" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> The work presented here makes contributions at three different levels. First, we have shown that potential impact of paraphrasing with respect to QA performance is significant. Replacing a question with a more felicitously worded question can potentially result in a 35% performance increase.</Paragraph>
    <Paragraph position="1"> Second, we performed our experiments by tapping into a readily available paraphrase resource: MT engines. Our results speak of the usefulness of the approach in producing paraphrases. This technique of obtaining a large, although low quality, set of paraphrases can be easily employed by other NLP practitioners wishing to investigate the impact of paraphrasing on their own problems.</Paragraph>
    <Paragraph position="2"> Third, we have shown that the task of selecting a better phrasing is amenable to learning, though more work is required to achieve its full potential. In that respect, the features and architecture discussed in Section 3 are a necessary first step in that direction. In future work, we are interested in developing effective filtering techniques to reduce our candidate set to a small number of high precision paraphrases, in experimenting with state-of-the-art paraphrasers, and in using paraphrasing to improve the stability of the QA system.</Paragraph>
  </Section>
class="xml-element"></Paper>