<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-1024">
  <Title>A Hybrid Approach to Natural Language Web Search</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> The popularity of natural language search is evidenced by the growing number of search engines, such as AskJeeves, Electric Knowledge, and Northern Light,1 that offer such functionality. For most sites, we were only able to perform a cursory examination of their proprietary techniques. Adopting a similar approach as FAQFinder (Hammond et al., 1995), AskJeeves maintains a database of questions and webpages that provide answers to them. User questions are compared against those in the database, and links to webpages for the closest matches are returned. Similar to our approach, Electric Knowledge transforms a natural language question into a series of increasingly more general key-word queries (Bierner, 2001). However, their query formulation process utilizes hard-crafted regular expressions, while we adopt a more general machine learning approach for transformation rule application. null Our work is also closely related to question answering in the question analysis component (e.g., (Harabagui et al., 2001; Prager et al., 2000; Clarke et al., 2001; Ittycheriah et al., 2001)). In particular, Harabagui et al.(2001) also iteratively reformulate queries based partly on the search results.</Paragraph>
    <Paragraph position="1"> However, their mechanism for query reformulation is heuristic-based. We utilized machine learning to</Paragraph>
  </Section>
class="xml-element"></Paper>