<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0425">
  <Title>Named Entity Recognition through Classifier Combination</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> This paper investigates the combination of a set of diverse statistical named entity classifiers, including a rule-based classifier - the transformation-based learning classifier (Brill, 1995; Florian and Ngai, 2001, henceforth fnTBL) with the forward-backward extension described in Florian (2002a), a hidden Markov model classifier (henceforth HMM), similar to the one described in Bikel et al. (1999), a robust risk minimization classifier, based on a regularized winnow method (Zhang et al., 2002) (henceforth RRM) and a maximum entropy classifier (Darroch and Ratcliff, 1972; Berger et al., 1996; Borthwick, 1999) (henceforth MaxEnt). This particular set of classifiers is diverse across multiple dimensions, making it suitable for combination: * fnTBL is a discriminant classifier - it bases its classification decision only on the few most discriminant features active on an example - while HMM, RRM and MaxEnt are agglomerative classifiers - their decision is based on the combination of all features active for the particular example.</Paragraph>
    <Paragraph position="1"> * In dealing with the data sparseness problem, fnTBL, MaxEnt and RRM investigate and integrate in their decision arbitrary feature types, while HMM is dependent on a prespecified back-off path.</Paragraph>
    <Paragraph position="2"> * The search methods employed by each classifier are different: the HMM, MaxEnt and RRM classifiers construct a model for each example and then rely on a sequence search such as the Viterbi algorithm (Viterbi, 1967) to identify the best overall sequence, while fnTBL starts with most frequent classification (usually per token), and then dynamically models the interaction between classifications, effectively performing the search at training time.</Paragraph>
    <Paragraph position="3"> * The classifiers also differ in their output: fnTBL and RRM return a single classification per example1, while the MaxEnt and HMM classifiers return a probability distribution.</Paragraph>
    <Paragraph position="4"> The remainder of the paper is organized as follows: Section 2 describes the features used by the classifiers, Section 3 briefly describes the algorithms used by each classifier, and Section 4 analyzes in detail the results obtained by each classifier and their combination.</Paragraph>
  </Section>
class="xml-element"></Paper>