<?xml version="1.0" standalone="yes"?>
<Paper uid="N03-1028">
  <Title>Shallow Parsing with Conditional Random Fields</Title>
  <Section position="7" start_page="2" end_page="2" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have shown that (log-)linear sequence labeling models trained discriminatively with general-purpose optimization methods are a simple, competitive solution to learning shallow parsers. These models combine the best features of generative nite-state models and discriminative (log-)linear classi ers, and do NP chunking as well as or better than ad hoc classi er combinations, which were the most accurate approach until now. In a longer version of this work we will also describe shallow parsing results for other phrase types. There is no reason why the same techniques cannot be used equally successfully for the other types or for other related tasks, such as POS tagging or named-entity recognition.</Paragraph>
    <Paragraph position="1"> On the machine-learning side, it would be interesting to generalize the ideas of large-margin classi cation to sequence models, strengthening the results of Collins (2002) and leading to new optimal training algorithms with stronger guarantees against over tting.</Paragraph>
    <Paragraph position="2"> On the application side, (log-)linear parsing models have the potential to supplant the currently dominant lexicalized PCFG models for parsing by allowing much richer feature sets and simpler smoothing, while avoiding the label bias problem that may have hindered earlier classi er-based parsers (Ratnaparkhi, 1997). However, work in that direction has so far addressed only parse reranking (Collins and Duffy, 2002; Riezler et al., 2002).</Paragraph>
    <Paragraph position="3"> Full discriminative parser training faces signi cant algorithmic challenges in the relationship between parsing alternatives and feature values (Geman and Johnson, 2002) and in computing feature expectations.</Paragraph>
  </Section>
class="xml-element"></Paper>