<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2422">
  <Title>Learning Transformation Rules for Semantic Role Labeling</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have described a Transformation-Based Error-Driven learning approach to the CoNLL shared task on semantic role labeling. Although we are relative newcomers to this task and this approach has not to our knowledge been applied to it before, we believe our results are of general interest for the following reasons.</Paragraph>
    <Paragraph position="1"> First, the learned output of the system is highly  scrutable, in the sense that the transformation rules can easily be reviewed and understood by a human supervisor. This may benefit real-world application of the technique as rules may be manually reordered, switched on or off, or modified. It also allows a developer to closely monitor changes in the system, creating new rules as he or she identifies areas of the data that are being underserved by the current set of transformation templates. Second, as alluded to above, there are several appealing directions to direct future research, and we believe the results obtained here can be significantly improved.</Paragraph>
    <Paragraph position="2"> Third, we know of no previous work using our look-behind reordering technique in conjunction with rule-based learning, and the technique may have broad applicability beyond semantic role labeling.</Paragraph>
  </Section>
class="xml-element"></Paper>