<?xml version="1.0" standalone="yes"?>
<Paper uid="H92-1093">
  <Title>Weight Estimation for N-Best Rescoring*</Title>
  <Section position="5" start_page="455" end_page="455" type="concl">
    <SectionTitle>
4. DISCUSSION
</SectionTitle>
    <Paragraph position="0"> In summary, we have described techniques that alleviate the problem of sensitivity to local optima in weight estimation for N-Best rescoring. However we find that there still exists a significant problem of mismatch between training and test sets. By comparing the contour plots we see that gender-independent optimization seems to be less sensitive to mismatch. This leads us to believe that we must estimate weights over a larger set of speakers.</Paragraph>
  </Section>
class="xml-element"></Paper>