Page 114 - Socially Intelligent Agents Creating Relationships with Computers and Robots

EgoChat Agent 97

virtualized-ego(A) virtualized-ego(B)
liquor brandy liquor
sake

beer wine

Figure 11.3. Example of a flow-of-topics set

Story structure: A stream of messages reflects social relationships between
community members. The frequency of exchange between community
members on mailing-lists and between a user and a VE on EgoChat
is recorded in each VE. For example, when community member (a) is
inclined to talk with member (b) more than with community member (c)
and the VE of member (b) has talked in the previous turn, the VE of
member (a), not that of member (c), goes first in the present turn.
Coherence: A VE that selects the following message after a message men-
tioned just before goes first so that messages are exchanged coherently.
The summaries in a personal memory are labeled with key words that
represent the contents of the summaries. A message is regarded as the
one that follows the previous message when its keyword matches the
previous one.
Fairness: A VE that speaks little goes before one that speaks a lot for the sake
of fairness.

4. Experiment

We carried out a basic experiment to ascertain the usability of the EgoChat
system and investigate the effects of voice interaction between humans and
agents. The experimental system was implemented in Java and Java3D API on
a MS-Windows OS. The voices of VEs were not generated by text-to-speech
software. Instead, they were recorded human voices. The body of a VE consists
of a spherical head and a cone-shaped torso, and the head nods while talking.

4.1 Method
We created four VEs and generated their personal memories from a humanly
summarized log of a mailing list about liquor where participants exchanged
ideas about how to market brandy. Each VE represents a mailing-list partici-
pant. The subjects were three postgraduate students in our laboratory who have