<?xml version="1.0" standalone="yes"?>
<Paper uid="A92-1002">
  <Title>A Dialog Control Algorithm and Its Performance</Title>
  <Section position="8" start_page="13" end_page="14" type="evalu">
    <SectionTitle>
6 Experimental Results
</SectionTitle>
    <Paragraph position="0"> The system has been implemented on a Sun 4 workstation with the majority of the code written in Quintus Prolog 1, and the parser in C. Speech recognition is performed by a Verbex 6000 running on an IBM PC and speech production is performed by a DECtalk 2 DTCO1 text-to-speech converter. The users are restricted to a  The implemented domain processor has been loaded with a model for a particular experimental circuit assembled on a Radio Shack 160-in-One Electronic Project Kit.</Paragraph>
    <Paragraph position="1"> After testing system prototypes with a few volunteers, eight subjects used the system during the formal experimental phase. After a warmup session where the sub-ject trained on the speech recognizer and practiced using the system, each subject participated in two sessions where up to ten problems were attempted. The system ran in declarative mode (user-controlled dialogs) during one session and in directive mode (strongly computer-controlled dialog) in the other session. Subjects attempted a total of 141 dialogs of which 118 or 84% were completed successfully. 3 Subjects spoke a total of 2840 user utterances, with 81.5% correctly interpreted by the system although only 50.0% were correctly recognized word for word by the speech recognizer. The average speech rate was 2.8 sentences per minute, and the average task completion times for successful dialogs were 4.5 and 8.5 minutes, respectively, for declarative and directive modes. The average number of user utterances per successful dialog was 10.7 in declarative mode and 27.6 in directive mode. A detailed description of the experiment and results is given in \[Smith, 1991\]. The substantially shorter completion times for users in declarative mode can be attributed to the fact that the subjects learned many of the debugging procedures during the experiment and did not need the detailed descriptions given in the directive mode.</Paragraph>
  </Section>
class="xml-element"></Paper>