<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-1054">
  <Title>A fast finite-state relaxation method for enforcing global constraints on sequence decoding</Title>
  <Section position="12" start_page="429" end_page="429" type="concl">
    <SectionTitle>
8 Conclusion
</SectionTitle>
    <Paragraph position="0"> Roth and Yih (2005) showed that global constraints can improve the output of sequence labeling models for semantic role labeling. In general, decoding under such constraints is NP-complete. We exhibited a practical approach, finite-state constraint relaxation, that greatly sped up decoding on this NLP task by using familiar finite-state operations--weighted FSA intersection and best-path extraction--rather than integer linear programming.</Paragraph>
    <Paragraph position="1"> We have also given a constraint relaxation algorithm for binary soft constraints. This allows incorporation of constraints akin to reranking features, in addition to inviolable constraints.</Paragraph>
  </Section>
class="xml-element"></Paper>