<?xml version="1.0" standalone="yes"?>
<Paper uid="H92-1009">
  <Title>Human-Machine Problem Solving Using Spoken Language Systems (SLS): Factors Affecting Performance and User Satisfaction</Title>
  <Section position="5" start_page="52" end_page="52" type="concl">
    <SectionTitle>
4. CONCLUSION
</SectionTitle>
    <Paragraph position="0"> Application development can benefit from analyses of factors affecting system performance and user satisfaction. We have presented examples of ways in which the behavior and satisfaction of subjects interacting with an SLS may be affected. We have described ways in which parameters of the system itself, such as speed and accuracy, affect different aspects of user satisfaction. We have examined the effect of user experience on recognition performance and found a decrease in word error rate over repeated scenarios.</Paragraph>
    <Paragraph position="1"> Adaptation was relatively greater for those subjects who had more than 20% errors on the first scenario. The decrease in errors could be attributed at least in part to a decrease in sentence perplexity and to a reduction in the use of out-of-vocabulary words. We have also shown a significant relationship between word error rates and hyperarticulation, a speech style that occurs relatively frequently with an imperfect recognizer. We have shown that instructions not to hyperarticulate reduced this maladaptive speech style, but that instructions did not result in improved recognition performance overall.</Paragraph>
    <Paragraph position="2"> Our studies have shown that along some dimensions, humans are flexible and can adapt in ways that improve system performance. However, hyperarticulation may be a maladaptive behavior for which a technological solution should be investigated. In particular we have found that strategies people use to try to improve normal human communication (e.g., hyperarticulation) can have the reverse effect in the context of our current models. While hyperarticulation is an &amp;quot;exaggerated&amp;quot; speech style that might improve comprehension for humans, it can cause poor recognition for automatic systems in which &amp;quot;exaggeration&amp;quot; is not adequately modeled.</Paragraph>
  </Section>
class="xml-element"></Paper>