<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-3002">
  <Title>Semantic Back-Pointers from Gesture</Title>
  <Section position="3" start_page="216" end_page="216" type="intro">
    <SectionTitle>
3 Future Directions
</SectionTitle>
    <Paragraph position="0"> The work on coreference can be considered preliminary, because it is focused on a subset of our corpus in which speakers use pre-printed diagrams as an explanatory aide. This changes their gestures (Eisenstein and Davis, 2003), increasing the proportion of deictic gestures, in which hand position is the most important feature (McNeill, 1992). Hand position is assumed to be less useful in characterizing the similarity of iconic gestures, which express meaning through motion or handshape. Using the subsection of the corpus in which no explanatory aids were provided, I will investigate how to assess the similarity of such dynamic gestures, in the hope that coreference resolution can still benefit from gestural cues in this more general case.</Paragraph>
    <Paragraph position="1"> Disfluency repair is another plausible domain in which gesture might improve performance. There are at least two ways in which gesture could be relevant to disfluency repair. Using the semantic back-pointer model, restart repairs could be identified if there is a strong gestural similarity between the original start point and the restart. Alternatively, gesture could play a pragmatic function, if there are characteristic gestures that indicate restarts or other repairs. In one case, we are looking for a similarity between the disfluency and the repair point; in the other case, we are looking for similarities across all disfluencies, or across all repair points. It is hoped that this research will not only improve processing of spoken natural language, but also enhance our understanding of how speakers use gesture to structure their discourse.</Paragraph>
  </Section>
class="xml-element"></Paper>