<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2055">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Analysis and Repair of Name Tagger Errors</Title>
  <Section position="4" start_page="0" end_page="420" type="intro">
    <SectionTitle>
2 Prior Work
</SectionTitle>
    <Paragraph position="0"> Some recent work has incorporated global information to improve the performance of name taggers. null For mixed case English data, name identification is relatively easy. Thus some researchers have focused on the more challenging task classifying names into correct types. In (Roth and  Yi 2002, 2004), given name boundaries in the text, separate classifiers are first trained for name classification and semantic relation detection. Then, the output of the classifiers is used as a conditional distribution given the observed data. This information, along with the constraints among the relations and entities (specific relations require specific classes of names), is used to make global inferences by linear programming for the most probable assignment. They obtained significant improvements in both name classification and relation detection.</Paragraph>
    <Paragraph position="1"> In (Ji and Grishman 2005) we generated N-best NE hypotheses and re-ranked them after coreference and semantic relation identification; we obtained a significant improvement in Chinese name tagging performance. In this paper we shall use a wider range of linguistic knowledge sources, and integrate cross-document techniques.</Paragraph>
  </Section>
class="xml-element"></Paper>