<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0831">
  <Title>Finding optimal parameter settings for high performance word sense disambiguation</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions. Further work.
</SectionTitle>
    <Paragraph position="0"> RLSC proved to be a very powerful learning model.</Paragraph>
    <Paragraph position="1"> We also believe that tuning the parameters of a model is a must, even if you have to invent parameters first. We think that the way we have proceeded here with a41 can be applied to other models, as a simple and direct post processing. Of course the right value of a41 has to be found case by case. We would suggest everyone who participated with systems that produce Bayesian-like class probabilities to try to apply this postprocessing to their systems.</Paragraph>
  </Section>
class="xml-element"></Paper>