<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2098">
  <Title>Exact Decoding for Jointly Labeling and Chunking Sequences</Title>
  <Section position="9" start_page="768" end_page="769" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have presented the decoding algorithm for label-chunk structure and showed its effectiveness in finding two layers of information, POS tags and NP chunks. This algorithm has a place between the  Viterbi algorithm for linear-chain models and the CKY algorithm for parsing, and the time complexity is O(n2). The use of our label-chunk structure significantly boosted the performance over cascaded CRFs despite the online learning algorithms used to train the system, and shows itself as a promising alternative to cascaded models, and possibly dynamic conditional random fields for modeling two layers of tags. Further work includes applying the algorithm to relation extraction, and devising an effective algorithm to find the marginal probabilities of parts.</Paragraph>
  </Section>
class="xml-element"></Paper>