<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1040">
  <Title>Expressing Implicit Semantic Relations without Supervision</Title>
  <Section position="10" start_page="318" end_page="319" type="evalu">
    <SectionTitle>
8 Conclusion
Latent Relational Analysis (Turney, 2005) pro-
</SectionTitle>
    <Paragraph position="0"> vides a way to measure the relational similarity between two word pairs, but it gives us little insight into how the two pairs are similar. In effect,  LRA is a black box. The main contribution of this paper is the idea of pertinence, which allows us to take an opaque measure of relational similarity and use it to find patterns that express the implicit semantic relations between two words.</Paragraph>
    <Paragraph position="1"> The experiments in Sections 5 and 6 show that ranking patterns by pertinence is superior to ranking them by a variety of tf-idf methods. On the word analogy and noun-modifier tasks, pertinence performs as well as the state-of-the-art, LRA, but pertinence goes beyond LRA by making relations explicit.</Paragraph>
  </Section>
class="xml-element"></Paper>