<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-1403">
  <Title>Towards Generation of Fluent Referring Action in Multimodal Situations</Title>
  <Section position="10" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Summary and future work
</SectionTitle>
    <Paragraph position="0"> What patterns of linguistic expressions are commonly used and how physical actions are temporally coordinated to them were reported based on cor8Tcl/tk is used for showing the GIF pictures and drawing the agent as well as other basic input/output functions; xanim is used for playing the digital movies.</Paragraph>
    <Paragraph position="1">  pus examinations. In particular, by categorizing objects according to two features, visibility and membership, the schemata of referring expressions could be derived. This categorization is still not sufficient for uniquely determining each reference expression, and some other features must impact the expressions used. This is, however, a good first step, as the two most dominant features were obtained. Moreover, the difference between the occurrence frequencies of those schemata in MMD and SMD explains the findings of our previous research. Implementation based on these results is on going.</Paragraph>
    <Paragraph position="2"> There is a lot of future work beyond the implementation issues. First, the reported coordination patterns between linguistic expressions and actions must be verified in a quantitative manner. An objective criterion for annotating visual information is hard to establish. Overcoming this problem is important and unavoidable. Next, our research must be generalized in two perspectives: the results must be confirmed in many materials other than our telephone; the degree of dependency on the language used must be examined.</Paragraph>
    <Paragraph position="3"> One of major problems stemming from our approach is that the importance of criteria is not clear. Although the criteria can be derived by observing and modeling the explanations made by experts, there may be fluent explanation techniques not yet observed. Deviations from the criteria do not cause a big problem, and the recipients do not perceive them to be awkward. These problems can be examined when the system currently implemented is made to generate several types of referring actions experimentally.</Paragraph>
  </Section>
class="xml-element"></Paper>