<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3702">
  <Title>Marianne.Starlander@eti.unige.ch</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Summary
</SectionTitle>
    <Paragraph position="0"> We have presented a first end-to-end evaluation of the MedSLT spoken language translation system. The medical students who tested it were all able to use the system well, with performance in some cases comparable to that of that of system developers after only two sessions. At least for the fairly simple type of diagnoses covered by our scenario, the system's performance appeared clearly adequate for the task.</Paragraph>
    <Paragraph position="1"> This is particularly encouraging, since the French to English version of the system is quite new, and has not yet received the level of attention required for a clinical system. The robustness added by the help system was sufficient to compensate for that, and in most cases, subjects were able to find ways to maneuver around coverage holes and other problems. It is entirely reasonable to hope that performance, which is already fairly good, would be substantially better with another couple of months of development work.</Paragraph>
    <Paragraph position="2"> In summary, we feel that this study shows that the conservative architecture we have chosen shows genuine potential for use in medical diagnosis situations. Before the end of 2006, we hope to have advanced to the stage where we can start initial trials with real doctors and patients.</Paragraph>
  </Section>
class="xml-element"></Paper>