<?xml version="1.0" standalone="yes"?>
<Paper uid="H94-1035">
  <Title>Information Based Intonation Synthesis*</Title>
  <Section position="7" start_page="195" end_page="196" type="concl">
    <SectionTitle>
5. Conclusions
</SectionTitle>
    <Paragraph position="0"> While previous attempts at intonation generation have relied on previous-mention heuristics for assigning accents, the present results show that is is possible to generate synthesized spoken responses with appropriate intonational contours in a database query task using explicit representations of contrastive stress. Many important problems remain, both because of the limited range of discourse-types and intonational tunes considered here, and because of the extreme oversimplification of the discourse model (particularly with respect to the  (14) Q: I know which procedure is recommended for the BURN patient,  but which procedure is recommended for the WOUND patient?  ontology, or variety of types of discourse entities). Nevertheless, the system presented here has a number of properties that we believe augur well for its extension to richer varieties of discourse, including the types of monologues and commentaries that are more appropriate for the actual TraumAID domain. Foremost among these is the fact that the system and the underlying theory are entirely modular. That is, any of its components can be replaced without affecting any other component because each is entirely independent of the particular grammar defined by the lexicon and the particular knowledge base that the discourse concerns. It is only because CCG allows us to unify the structures implicated in syntax and semantics on the one hand, and intonation and discourse information on the other, that this modular structure can be so simply attained.</Paragraph>
  </Section>
class="xml-element"></Paper>