<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-1206">
  <Title>Computing prosodic properties in a data-to-speech system</Title>
  <Section position="8" start_page="43" end_page="43" type="evalu">
    <SectionTitle>
5 Evaluation
</SectionTitle>
    <Paragraph position="0"> Nachtegaal (1997) reports on a small experiment which was carried out to test the accentuation algorithm of D2S. No formal evaluation has taken place for the algorithm determining the placement of phrase boundaries.</Paragraph>
    <Paragraph position="1"> hL the experiment by Nachtegaal (1997), Dutch speakers were asked to read aloud texts generated by the LGM of GoalGetter. Recordings of the read texts were presented to 'expert listeners' who indicated on which words they heard an accent. Comparisons were then made between the accentuation patterns produced by the speakers and those generated by the system. The results of the experiment were positive: the number of words on which the accentuation by the speakers deviated from the accentuation by GoalGetter was very small (less than 470 of all accentable words, i.e., excluding 'unaccentables' like function words etc.). The texts used in the experiment contained sentences which were structurally similar to those of the example text given in  currently generated by GoalGetter were included in the test. The prosody of the current version of Goal Getter was only evaluated informally, but the results were in line with those of Nachtegaal (1997).</Paragraph>
    <Paragraph position="2"> The prosodic rules described in this paper have also been succesfully implemented in the DYDsystem (I:LEF), which differs from GoalGetter with respect to language (English versus Dutch) and domain (Mozart compositions versus football reports).</Paragraph>
    <Paragraph position="3"> Informal evaluation of the prosody in DYD gave equally satisfactory results as for GoalGetter. This was as expected, since the prosodic rules of D2S are essentially domain- and language independent. 6 All things considered, the quality of the prosodic rules of D2S is judged to be good.</Paragraph>
  </Section>
class="xml-element"></Paper>