<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2022">
  <Title>Automatically Extracting Nominal Mentions of Events with a Bootstrapped Probabilistic Classifier[?]</Title>
  <Section position="6" start_page="173" end_page="173" type="relat">
    <SectionTitle>
4 Related Work
</SectionTitle>
    <Paragraph position="0"> In recent years an array of new approaches have been developed using weakly-supervised techniques to train classifiers or learn lexical classes or synonyms, e.g. (Mihalcea, 2003; Riloff and Wiebe, 2003). Severalapproachesmakeuseofdependency triples (Lin, 1998; Gorman and Curran, 2005). Our vector representation of the behavior of a word type across all its instances in a corpus is based on Lin (1998)'s DESCRIPTION OF A WORD.</Paragraph>
    <Paragraph position="1"> Yarowsky (1995) uses a conceptually similar technique for WSD that learns from a small set of seed examples and then increases recall by bootstrapping, evaluated on 12 idiosyncratically polysemous words. In that task, often a single disambiguating feature can be found in the context of a polysemous word instance, motivating his use of the decision list algorithm. In contrast, the goal here is to learn how event-like or non-event-like a set of contextual features together are. We do not expect that many individual features correlate unambiguously with references to events (or nonevents), only that the presence of certain features make an event interpretation more or less likely.</Paragraph>
    <Paragraph position="2"> This justifies our probabilistic Bayesian approach, which performs well given its simplicity.</Paragraph>
    <Paragraph position="3">  ThelenandRiloff(2002)useabootstrappingalgorithm to learn semantic lexicons of nouns for six semantic categories, one of which is EVENTS.</Paragraph>
    <Paragraph position="4"> For events, only 27% of the 1,000 learned words are correct. Their experiments were on a much smaller scale, however, using the 1,700 document MUC-4 data as a training corpus and using only 10 seeds per category.</Paragraph>
    <Paragraph position="5"> Most prior work on event nominals does not try to classify them as events or non-events, but instead focuses on labeling the argument roles based on extrapolating information about the argument structure of the verbal root (Dahl et al., 1987; Lapata, 2002; Pradhan et al., 2004). Meyers, et al.</Paragraph>
    <Paragraph position="6"> (1998) describe how to extend a tool for extraction of verb-based events to corresponding nominalizations. Hull and Gomez (1996) design a set of rule-based algorithms to determine the sense of a nominalization and identify its arguments.</Paragraph>
  </Section>
class="xml-element"></Paper>