<?xml version="1.0" standalone="yes"?>
<Paper uid="W96-0418">
  <Title>Matchmaking: dialogue modelling and speech generation meet*</Title>
  <Section position="7" start_page="176" end_page="179" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> In this article we have developed a model for guiding the selection of intonation in a system supporting human-machine interaction in retrieval dialogues with spoken output. We have concentrated on the choice \[1\] of tone as a major signal of interpersonal semantic features, such as speech acts and speaker's attitudes.</Paragraph>
    <Paragraph position="1"> To express these appropriately is crucial especially in human-machine dialogue, since they contribute to the \[2\] success of the interaction in a major way.</Paragraph>
    <Paragraph position="2"> As a starting point we have taken two existing \[3\] systems--the COR dialogue model and the KOMET-PENMAN generation system. On this basis, we have determined a number of factors that contribute to the selection of appropriate tones, such as speech func- \[4\] tion, speaker's attitudes and hearer's expectations, and types of dialogue moves in context. Finally, we have proposed a stratified model that includes all of the relevant kinds of information to guide the selection of tone. \[5\] Even though our dialogue model was originally not designed for language, we have shown that this relatively \[6\] simple model provides useful information for intonation selection. A linguistically based discourse model would be able to provide more information, but in the context of an interactive conversational system in which there \[7\] are practical limits on how tong it can take to produce a response, we believe that a full fledged discourse analysis system would be too slow.</Paragraph>
    <Paragraph position="3"> We are aware that we have left untouched a number of \[8\] problems that are involved in the generation of appropriate intonations. These include: \[9\] * accounting for the textual meaning of intonation encoded in information structure and thematic  development/progression (realized in tonicity; see Section 2); We handle only situations in which there is a one-to-one corresponds between tone group and clause.</Paragraph>
    <Paragraph position="4"> We can only make predictions about complete clauses, hence the grammar prevents the generation of utterances with ellipses. This is relevant for geographical clarification question, e.g., &amp;quot;Wollen Sie nach Frankfurt am Main oder Frankfurt an der Oder?&amp;quot;. In many contexts it is more natural to use just a phrase &amp;quot;Frankfurt am Main oder an der Oder?&amp;quot; Similarly it is unnatural to generate the evaluate moves as complete clauses. It suffices to generate simple phrases like &amp;quot;Thanks&amp;quot; or &amp;quot;OK&amp;quot;. Also, the method we have applied here to develop our model has been solely qualitative. For a proper validation we need to analyse larger quantities of dialogue in order to have an empirically sound foundation. The same is true for the classification of intonation which was developed by Pheby in the late sixties. Here, the collaborative work with speech synthesis will provide us with empirical data that can then be used to refine the classification.</Paragraph>
  </Section>
class="xml-element"></Paper>