<?xml version="1.0" standalone="yes"?>
<Paper uid="H89-2055">
  <Title>COORDINATING TEXT AND GRAPHICS IN EXPLANATION GENERATION</Title>
  <Section position="8" start_page="430" end_page="431" type="concl">
    <SectionTitle>
6 CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> We have focused on three features of COMET's architecture that allow the dynamic generation of integrated multimedia explanations: a common content description, the fine-grained assignment of information to media, and bidirectional interaction among components. The use of an annotated common content description allows each  media-specific generator to be aware of all information to be communicated in the explanation, and to use that information to influence the realization of segments for which it is responsible. Our media coordinator allows for small portions of the same LF to be realized in different media. For example, location modifiers of an object may be expressed in graphics only, while the remainder of the LF is expressed in both text and graphics. Similarly, conditionals may be expressed in text only, while the conjoined actions may be expressed in both text and graphics. Finally, our proposed approach for accomplishing bidirectional interaction between components will make it possible for the text and graphics generators to communicate with other. This will allow decisions made by each generator to influence the other.</Paragraph>
    <Paragraph position="1"> We are pursuing a number of different research directions, in addition to our work on bidirectional interaction. Our media coordinator is currently more of a dictator than a coordinator. We are interested in developing strategies for those situations in which the media generators determine that the assignments made by the media coordinator are unsatisfactory. In these cases, the generators could provide feedback to the coordinator, which could in turn modify its plan. We are also interested in situations where context influences the selection of media. Finally, although a single media-independent content planner has definite advantages, there are situations in which it, too, should accept feedback from the generators and modify the content specification.</Paragraph>
  </Section>
class="xml-element"></Paper>