<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1015">
  <Title>Espresso: Leveraging Generic Patterns for Automatically Harvesting Semantic Relations</Title>
  <Section position="12" start_page="119" end_page="119" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> We proposed a weakly-supervised, generalpurpose, and accurate algorithm, called Espresso, for harvesting binary semantic relations from raw text. The main contributions are: i) a method for exploiting generic patterns by filtering incorrect instances using the Web; and ii) a principled measure of pattern and instance reliability enabling the filtering algorithm.</Paragraph>
    <Paragraph position="1"> We have empirically compared Espresso's precision and recall with other systems on both a small domain-specific textbook and on a larger corpus of general news, and have extracted several standard and specific semantic relations: isa, part-of, succession, reaction, and production.</Paragraph>
    <Paragraph position="2"> Espresso achieves higher and more balanced performance than other state of the art systems. By exploiting generic patterns, system recall substantially increases with little effect on precision. There are many avenues of future work both in improving system performance and making use of the relations in applications like question answering. For the former, we plan to investigate the use of WordNet to automatically learn selectional constraints on generic patterns, as proposed by (Girju et al. 2006). We expect here that negative instances will play a key role in determining the selectional restrictions.</Paragraph>
    <Paragraph position="3"> Espresso is the first system, to our knowledge, to emphasize concurrently performance, minimal supervision, breadth, and generality. It remains to be seen whether one could enrich existing ontologies with relations harvested by Espresso, and it is our hope that these relations will benefit NLP applications.</Paragraph>
  </Section>
class="xml-element"></Paper>