MIT Media Laboratory
Room E15-320R, 20 Ames St.
Cambridge,
MA 02139-4307
Email: hannes@media.mit.edu
Research Summary
In my work, I am applying theory of human communicative
behavior to the design of autonomously animated avatars. In particular I
am interested in how fundamental conversational functions such as displaying
awareness and attention, establishing and sustaining an open channel, signaling
intention and synchronizing turns are realized nonverbally. These are all
functions that we naturally and spontaneously employ in face-to-face
conversation, but are lost when we introduce avatars that depend on explicit
control. My approach is to implement the avatar as an animated agent,
capable of autonomously carrying out spontaneous conversational functions
nonverbally based on the social situation and the user's communicative
intention. This work is a part of the research in the Gesture and Narrative Language
Group lead by professor Justine
Cassell.
Research Methodology
The research process cycles through four important levels:
research existing theory, expand theory through empirical studies, implement a
computational model in an avatar and finally evaluate the avatar in use.
At the theoretical level I am drawing from the rich literature in social
interaction and discourse analysis. We also conduct experiments when
current theory falls short of explaining a particular phenomenon. In those
experiments we observe and analyze naturally occurring human-human interaction,
paying particular attention to the relationship between the verbal and the
non-verbal. The implementation of new avatar behaviors and the evaluation
of their impact on online communication are carried out using a prototype
system. The evaluation involves rigorous user testing where
quantitative and qualitative measurements are used to judge factors such as
user's preception of naturalness, expressivity and conversational control.
Future Directions for My Research
- Group Interaction. So far I have concentrated on the interaction
among pairs of users. However, studying phenomena such as group-formation,
turn-taking and side conversations in a group discussion is a natural next
step.
- Agent Architecture. As the avatars become more complex
and exhibit more autonomy and understanding, their internal architecture
becomes an increasingly interesting issue. Are traditional agent
architectures directly applicable? How is the presence of the user captured
in the architecture?
- User Interface. Although my approach saves the user from having
to micromanage avatar behavior, the user's high level intentions still have
to be derived from user input. What are the affordances of traditional
input devices? What about other modalities such as audio or vision?
- Community Impact. All evaluation is currently
done using a prototype system and the test subjects only get a chance to use
it for a short period of time. To truly measure the contribution of my
research to online communities, long term studies with a large user base
must be performed. To this end, I hope to release a publicly available
version of the client software.
Other Important Issues for the Field
- Character Animation. Once the desired behavior has been chosen,
the avatar degrees of freedom have to be animated in a life-like fashion.
There is a variety of approaches to choose from, ranging from
motion-capture-based 3D animation to pre-drawn 2D sprites. How do these
approaches lend themselves to the successful animation of communicative
behavior? What new techniques are called for?
- Computer Mediated Communication. Avatar based communication
is a part of the larger field of Computer Mediated Communication where
considerable effort has been put into enhancing video conferencing, network
telephony and shared desktops. How do avatars compare to these other methods
for transmitting presence? What about the difference between graphically
mediated presence and textual representation (such as found in MUDs)?
- Virtual World Design. Clearly the design of the
environment in which the avatars live has an effect on the communication
taking place. How aware of the environment does the avatar have to be? How
do different social spaces, such as common vs. private spaces, influence
communicative behavior? How does the move from true-to-life worlds to more
abstract worlds affect avatar design?
Selected Publications
- "Autonomous Communicative Behaviors in Avatars"
(abstract). Proceedings of Lifelike Computer Characters '96,
Utah,
October 8-11, p.49., 1996 [PS]
[PDF]
- "Autonomous Communicative Behaviors in Avatars." Master's
Thesis in Media Arts and Sciences. MIT Media Laboratory 1997. [HTML]
- "BodyChat: Autonomous Communicative Behaviors in Avatars."
With Justine Cassell.
Proceedings of ACM Autonomous Agents '98, Minneapolis, May 9-13, p.269-276.,
1998 [PS]
[PDF]
- "Fully Embodied Conversational Avatars: making Communicative
Behaviors Autonomous." With Justine Cassell (first author).
Autonomous Agents and Multi-Agent Systems Journal (2). ACM. (to appear)