Graphics Reference
In-Depth Information
conditions at the target location. In other words, the suggestive
presentation in the initial legend of a presumable summer holiday of
fourteen days had worked well and, after being informed that Waiuku
is in the southern hemisphere, most subjects tried hard to readjust the
contents of their suitcases. This was clearly expressed in the answers
of those subjects with interventions as well in the answers to the final
questions by all the subjects.
The role of the computer voice
We had specifically chosen a computer voice for the experiment that
sounded more technical and machine-like than the state-of-the-art
text-to-speech (TTS) systems. This decision was deliberate and was
intended to support the illusion of interacting with a technical system.
As the reactions of the subjects showed, this rationale proved to be
successful. We could not find any indications that the subjects doubted
that they were interacting with a technical device. Those subjects who
referred to the quality of the computer voice during the interviews
consistently emphasized voice quality as being highly relevant for
their subjective impressions of the interaction.
Challenges for the subjects
During their initial briefing, the subjects were informed that all
interactions would be based on speech only, and that neither a
keyboard nor a mouse would be available to them. Since the briefing
did not comprise any detailed information about the natural language
processing and the problem-solving capabilities or limitations of
the system, the subjects were more or less forced to actively explore
these aspects during the course of the interaction. The challenge
for the subjects was twofold: They had to find out how (i.e. with
which actions) they could solve problems that they encounter during
interaction, and also to discover which linguistic means were available
for them to instruct the system to perform the necessary actions. In
other words, in order to be successful, they had to build a model of
the capacities and limitations of the system based on their experiences
of successful and unsuccessful interactions. The user's model of the
system would, of course, strongly influence the user's behavior and
the subsequent course of the interaction.
4.3 Sample
The generated LAST MINUTE corpus was based on 135 experiments.
Two experiments were aborted due to technical problems; three
Search WWH ::




Custom Search