<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-0306">
  <Title>A Transformational-based Learner for Dependency Grammars in Discharge Summaries</Title>
  <Section position="9" start_page="2" end_page="2" type="concl">
    <SectionTitle>
8 Conclusions
</SectionTitle>
    <Paragraph position="0"> Natural language processors in the medical domain will be more flexible and portable with assisted lexicon design. The syntactic dependencies in a dependency grammar may be useful for the lexical acquisition necessary to make this possible. We have investigated using transformational-based learning as a technique for learning a dependency grammar in a medical corpus. To better learn dependency grammars we used a template design which uses the structure of the parse tree explicitly and transformations that operated directly on the trees. Training on a set of 830 sentences of parsed medical discharge summaries gave a best parser with 77% accuracy. The inclusion of tree information in the template design slightly improved the parser. The rules produced were intuitive and understandable, and the limited amount of training material will allow the technique to be used on other medical domains without extensive manual parsing. Further work will test the utility of head-dependency relationships for machine learning semantic classes.</Paragraph>
  </Section>
class="xml-element"></Paper>