<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0612">
  <Title>An Expectation Maximization Approach to Pronoun Resolution</Title>
  <Section position="4" start_page="0" end_page="88" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> Pronoun resolution typically employs some combination of constraints and preferences to select the antecedent from preceding noun phrase candidates. Constraints filter the candidate list of improbable antecedents, while preferences encourage selection of antecedents that are more recent, frequent, etc. Implementation of constraints and preferences can be based on empirical insight (Lappin and Leass, 1994), or machine learning from a reference- null annotated corpus (Ge et al., 1998). The majority of pronoun resolution approaches have thus far relied on manual intervention in the resolution process, such as using a manually-parsed corpus, or manually removing difficult non-anaphoric cases; we follow Mitkov et al.'s approach (2002) with a fully-automatic pronoun resolution method. Parsing, noun-phrase identification, and non-anaphoric pronoun removal are all done automatically.</Paragraph>
    <Paragraph position="1"> Machine-learned, fully-automatic systems are more common in noun phrase coreference resolution, where the method of choice has been decision trees (Soon et al., 2001; Ng and Cardie, 2002). These systems generally handle pronouns as a subset of all noun phrases, but with limited features compared to systems devoted solely to pronouns. Kehler used Maximum Entropy to assign a probability distribution over possible noun phrase coreference relationships (1997). Like his approach, our system does not make hard coreference decisions, but returns a distribution over candidates.</Paragraph>
    <Paragraph position="2"> The above learning approaches require annotated training data for supervised learning. Cardie and Wagstaff developed an unsupervised approach that partitions noun phrases into coreferent groups through clustering (1999). However, the partitions they generate for a particular document are not useful for processing new documents, while our approach learns distributions that can be used on unseen data. There are also approaches to anaphora resolution using unsupervised methods to extract useful information, such as gender and number (Ge et al., 1998), or contextual role-knowledge (Bean and Riloff, 2004). Co-training can also leverage unlabeled data through weakly-supervised reference resolution learning (M&amp;quot;uller et al., 2002). As an alternative to co-training, Ng and Cardie (2003) use EM to augment a supervised coreference system with unlabeled data. Their feature set is quite different, as it is designed to generalize from the data in a labeled set, while our system models individual words. We suspect that the two approaches can be combined.</Paragraph>
    <Paragraph position="3"> Our approach is inspired by the use of EM in bilingual word alignment, which finds word-to-word correspondences between a sentence and its translation. The prominent statistical methods in this field are unsupervised. Our methods are most influenced by IBM's Model 1 (Brown et al., 1993).</Paragraph>
  </Section>
class="xml-element"></Paper>