<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0607">
  <Title>Feeding OWL: Extracting and Representing the Content of Pathology Reports</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Summary and Further Work
</SectionTitle>
    <Paragraph position="0"> We have described LUPUS, an NLP system that makes use of a domain ontology to guide extraction of information about entities from medical texts, and represents this information as instances of concepts from that ontology. Besides its direct use for content-based search on these texts, the fact that the system relies entirely on emerging semantic web standards will make the resulting annotated information usable for all kinds of agents working with such data.</Paragraph>
    <Paragraph position="1"> As a next step, we plan to add discourse processing to the pipeline (see e.g. (Hahn et al., 1998) for a discussion why such a step is required even for such relatively simple texts). As mentioned above, the prerequisite information (about definite articles, for example) is already there; we plan to use the available domain knowledge to guide the search for antecedents for bridging. As a more technical improvement we are investigating ways of making the architecture less pipeline-y, and to integrate domain reasoning in computing edges in the chart. Lastly, we are also working on a large-scale evaluation of the system, by manually annotating reports to compute precision and recall.</Paragraph>
  </Section>
class="xml-element"></Paper>