<?xml version="1.0" standalone="yes"?>
<Paper uid="P05-1021">
  <Title>Improving Pronoun Resolution Using Statistics-Based Semantic Compatibility Information</Title>
  <Section position="3" start_page="0" end_page="165" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Semantic compatibility is an important factor for pronoun resolution. Since pronouns, especially neutral pronouns, carry little semantics of their own, the compatibility between an anaphor and its antecedent candidate is commonly evaluated by examining the relationships between the candidate and the anaphor's context, based on the statistics that the corresponding predicate-argument tuples occur in a particular large corpus. Consider the example given in the work of Dagan and Itai (1990): (1) They know full well that companies held tax money aside for collection later on the basis that the government said it1 was going to collect it2.</Paragraph>
    <Paragraph position="1"> For anaphor it1, the candidate government should have higher semantic compatibility than money because government collect is supposed to occur more frequently than money collect in a large corpus. A similar pattern could also be observed for it2.</Paragraph>
    <Paragraph position="2"> So far, the corpus-based semantic knowledge has been successfully employed in several anaphora resolution systems. Dagan and Itai (1990) proposed a heuristics-based approach to pronoun resolution. It determined the preference of candidates based on predicate-argument frequencies. Recently, Bean and Riloff (2004) presented an unsupervised approach to coreference resolution, which mined the co-referring NP pairs with similar predicate-arguments from a large corpus using a bootstrapping method.</Paragraph>
    <Paragraph position="3"> However, the utility of the corpus-based semantics for pronoun resolution is often argued.</Paragraph>
    <Paragraph position="4"> Kehler et al. (2004), for example, explored the usage of the corpus-based statistics in supervised learning based systems, and found that such information did not produce apparent improvement for the overall pronoun resolution. Indeed, existing learning-based approaches to anaphor resolution have performed reasonably well using limited and shallow knowledge (e.g., Mitkov (1998), Soon et al. (2001), Strube and Muller (2003)).</Paragraph>
    <Paragraph position="5"> Could the relatively noisy semantic knowledge give us further system improvement? In this paper we focus on improving pronominal anaphora resolution using automatically computed semantic compatibility information. We propose to enhance the utility of the statistics-based knowledge from two aspects: Statistics source. Corpus-based knowledge usually suffers from data sparseness problem. That is, many predicate-argument tuples would be unseen even in a large corpus. A possible solution is the  web. It is believed that the size of the web is thousands of times larger than normal large corpora, and the counts obtained from the web are highly correlated with the counts from large balanced corpora for predicate-argument bi-grams (Keller and Lapata, 2003). So far the web has been utilized in nominal anaphora resolution (Modjeska et al., 2003; Poesio et al., 2004) to determine the semantic relation between an anaphor and candidate pair. However, to our knowledge, using the web to help pronoun resolution still remains unexplored.</Paragraph>
    <Paragraph position="6"> Learning framework. Commonly, the predicate-argument statistics is incorporated into anaphora resolution systems as a feature. What kind of learning framework is suitable for this feature? Previous approaches to anaphora resolution adopt the single-candidate model, in which the resolution is done on an anaphor and one candidate at a time (Soon et al., 2001; Ng and Cardie, 2002). However, as the purpose of the predicate-argument statistics is to evaluate the preference of the candidates in semantics, it is possible that the statistics-based semantic feature could be more effectively applied in the twin-candidate (Yang et al., 2003) that focusses on the preference relationships among candidates.</Paragraph>
    <Paragraph position="7"> In our work we explore the acquisition of the semantic compatibility information from the corpus and the web, and the incorporation of such semantic information in the single-candidate model and the twin-candidate model. We systematically evaluate the combinations of different statistics sources and learning frameworks in terms of their effectiveness in helping the resolution. Results on the MUC data set show that for neutral pronoun resolution in which an anaphor has no specific semantic category, the web-based semantic information would be the most effective when applied in the twin-candidate model: Not only could such a system significantly improve the baseline without the semantic feature, it also out-performs the system with the combination of the corpus and the single-candidate model (by 11.5% success). null The rest of this paper is organized as follows. Section 2 describes the acquisition of the semantic compatibility information from the corpus and the web. Section 3 discusses the application of the statistics in the single-candidate and twin-candidate learning models. Section 4 gives the experimental results, and finally, Section 5 gives the conclusion.</Paragraph>
  </Section>
class="xml-element"></Paper>