<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-3004">
  <Title>Virtual Modality: a Framework for Testing and Building Multimodal Applications</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Summary and Future Work
</SectionTitle>
    <Paragraph position="0"> An experimental framework has been introduced, called Virtual Modality, which aims to assist in the development and testing of multimodal systems. The paper explained the motivation behind generating (simulated) multimodal data from available textual representations of logged user inputs. The procedure of replacing loca-tion references with deictic expressions, as well as the placement of the referenced semantic value along with temporal information to the Virtual Modality channel were explained. Lastly, the evaluation scenario and preliminary test results were presented.</Paragraph>
    <Paragraph position="1"> Future work will investigate the following topics: * how the generated Virtual Modality data can be utilized to train a statistical (or hybrid) multimodal integration module; * how adaptivity in the integration module can be achieved using a vast amount of training data; * how N-best choices in the Virtual Modality input can be utilized; * whether disambiguation can be achieved with training examples covering all temporal cases; * how evidence in one modality can help to resolve some ambiguity in the other modality, and ultimately, how to provide an accurate interpretation of the overall user intention.</Paragraph>
  </Section>
class="xml-element"></Paper>