<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1655">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics A Hybrid Markov/Semi-Markov Conditional Random Field for Sequence Segmentation</Title>
  <Section position="8" start_page="471" end_page="471" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have shown that order-1 semi-Markov conditional random fields are strictly more expressive than order-1 Markov CRFs, and that the added expressivity enables the use of features that lead to improvements on a segmentation task. On the other hand, Markov CRFs can more naturally incorporate certain features that may be useful for modeling sub-chunk phenomena and generalization to unseen chunks. To achieve the best performance for segmentation, we propose that both types of features be used, and we show how this can be done efficiently.</Paragraph>
    <Paragraph position="1"> Additionally, we have shown that a log conditional odds feature estimated from a generative model can be superior to the more common log conditional probability.</Paragraph>
  </Section>
class="xml-element"></Paper>