Text size
  • Small
  • Medium
  • Large
Contrast
  • Standard
  • Blue text on blue
  • High contrast (Yellow text on black)
  • Blue text on beige

    Wizard of Oz Experiments and Companion Dialogues

    People and Computers XXIV Games are a Serious Business

    Proceedings of HCI 2010
    The 24th British HCI Group Annual Conference
    University of Abertay, Dundee, UK

    6 - 10 September 2010

    AUTHORS

    Jay Bradley, David Benyon, Oli Mival and Nick Webb

    ABSTRACT

    Novel speech systems such as the conversational agents being developed by the Companions Project (www.companions-project.org) can be simulated using the Wizard of Oz methodology. In this approach technologies that are not yet ready for testing by people are replaced by a human, both for prototyping and collecting additional dialogue data. In the case of Companions we want to observe what it would be like for people interact with a fully functional embodied conversational agent (ECA) and to collect samples of typical dialogue in order to explore, evaluate and model dialogue strategies. One controversial aspect of the Wizard of Oz approach is whether people should be aware that they are interacting with a simulation or whether they should be “fooled” into thinking they are interacting with a real system. Clearly there are ethical issues involved in fooling people, but some argue that unless the participant believes the simulation to be real, the results of any experimentation will not be applicable to the real situation. Over the course of several previous Wizard of Oz experiments our observations suggest that the dialogues produced do not significantly differ whether the participants know that the technology is faked or not. This hypothesis was investigated by collecting dialogues from two groups of participants. One group of participants believed that the Wizard of Oz speech system was in fact a fully computerised prototype and the other group knew that they would be talking through the interface to a hidden person (the wizard). The dialogues were analysed for differences attributable to the participants’ beliefs about the system. This analysis was undertaken by an independent “blind” reviewer, a dialogue expert who attempted to allocate participants to one group or the other. His guess was wrong for four out of the six participants. Thus it appears that whether people believe they are interacting with a real system or not does not effect the dialogues and other factors, for example the personality of the person engaged in a dialogue with an ECA, are more important.

    PAPER FORMATS

    PDF filePDF Version of this Paper (709kb)