<?xml version="1.0" standalone="yes"?>
<Paper uid="H94-1074">
  <Title>Speech-Based Retrieval Using Semantic Co-Occurrence Filtering</Title>
  <Section position="5" start_page="375" end_page="375" type="evalu">
    <SectionTitle>
5. Evaluation
</SectionTitle>
    <Paragraph position="0"> We created 100 queries, each composed of a few words that characterize a topic of interest (e.g. =Apollo space program&amp;quot;).</Paragraph>
    <Paragraph position="1"> To evaluate the benefit of semantic co-occurrence filtering directly, we verified that the words we selected had entries in the phonetic dictionary and that the encyclopedia contained at least one relevant article.</Paragraph>
    <Paragraph position="2"> Of the 100 queries, 83 were successful (i.e. retrieved at least one relewnt article in the top 25 titles). If only the top word hypotheses from the n-best component were inserted in the boolean queries, only 32 of the queries would succeed. The In constructing the test queries, sometimes only a single word immediately came to mind for some topics. In such cases we found that a useful strategy for adding another word was to use either a name, hyponym or hypernym. Thus the word =ant&amp;quot; was augmented by adding =insect&amp;quot; as a second word. Although less robust, single word queries are not precluded.</Paragraph>
    <Paragraph position="3"> Either their length may distinguish them (e.g. =savonarola&amp;quot; and =nitroglycerin&amp;quot;) or the IR query can be constructed by duplicating the OR term for the single word (the constraint is then word recurrence which still has value for filtering).</Paragraph>
  </Section>
class="xml-element"></Paper>