<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-2136">
  <Title>Confirmation in Multimodal Systems</Title>
  <Section position="9" start_page="827" end_page="827" type="concl">
    <SectionTitle>
7 CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> We have developed a system that meets the following expectation: when the proposition being confirmed is a command, it should be one that the system believes can be executed. To meet this expectation and increase the conversational performance of multimodal systems, we have argued that confirmations should occur late in the system's understanding process, at a point after blending has enhanced its understanding. This research has compared two strategies: one in which confirmation is performed immediately after speech recognition, and one in which it is delayed until after multimodal integration. The comparison shows that late confirmation reduces the time to perform map manipulation tasks with a multimodal interface. Users can interact faster and complete commands in fewer tums, leading to a reduction in collaborative effort.</Paragraph>
    <Paragraph position="1"> A direction for future research is to adopt a strategy for determining whether a confirmation is necessary \[29, 30\], rather than confu'rning every utterance, and measuring this strategy's effectiveness.</Paragraph>
  </Section>
class="xml-element"></Paper>