Embodied Conversational Agents are virtual agents that have anthropometric features and that display a multimodal behavior. In particular, they are able to perceive and understand what a user is saying as well as to converse with him or another agent using verbal and nonverbal behaviors. They may talk, smile, nod, gestures...

Their physical appearance may greatly vary a lot: 2D vs 3D, realistic vs cartoonish, full-body vs face-only. Moreover these agents may intervene with the virtual environment they are placed in; they may look around, look at objects, grab them, move them around, and so on.

These agents are getting very popular in their use as interface. The user may dialog with them to get information or to have an action performed. Endowing agents with communicative and expressive behavior often increases the impact of the ECA on the interaction. Several applications may benefit from them; several pedagogical agent have been proposed in E-learning systems; they may be used for entertainment or for telecommunication applications; many web-agents that help the user surfing the web pages of a company have been proposed.

The aim of this tutorial is to define the many aspects involved in the creation of ECAs. The tutorial will also outline the theoretical fundations behind such a creation. In particular it will detail the link between intonation and nonverbal behavior, the synchronization scheme between the verbal and nonverbal behavior, model of communicative functions, of social behavior, of gesture coordination and expressivity.

It will also present ways of representing actions and nonverbal behaviors to drive the animation of the agent. The tutorial will look at the several type of embodiement and appearance of an agent and present models and representation language that are adequate for the different types. In each of these themes implementation details will be provided. The tutorial will also describe existing tools to generate and manipulate ECA's behavior. Its aim is, thus, to provide a broad view on the ECA, pointing out and giving information on the different aspects and their corresponding problems of ECA.

Course Outline:

9:00 - 10:00  LINGUISTIC VALUE (Doug DeCarlo - Matthew Stone) 
              meaning / signal (theory) 
	      language of representation 
              Ruth (tools)


10:00-10:45   COMMUNICATIVE FUNCTIONS (Catherine Pelachaud) 
              taxonomy of communicative function 
              language of representation (APML) 
              Greta (tools)  

10:45-11:15   Break

11:15-12:00   GAZE BEHAVIOR (Norman Badler - Catherine Pelachaud) 
              communicative model 
              statistical model 
              attention model 

12:00-12:30   CARTOON and CARICATURE (Kris Thorisson) 
              Facial Animation Specification 
              design goals  

12:00-1:30    Lunch

1:30-2:30     MULTI-AGENT INTERACTION (Thomas Rist) 
	      model of social context 
              dialog generation 
              arena server (application)  

2:30- 3:30    GESTURE EXPRESSIVITY (Norman Badler) 
              Laban Annotation Movement 
	      Expressive gesture 
              EMOTE (tools) 

3:30-4:00     Break

4:00-5:00     GESTURE COORDINATION (Stephane Donikian) 
              dynamic behavior 
              perception-decision-action 
              programming environment 
              applications

5:00-5:45     ARCHITECTURES FOR COMMUNICATIVE HUMANOIDS (Kris Thorisson) 
              an integrated system for developing 
              interactive communicative humanoids

5:45-6:00     CONCLUSION AND DISCUSSION

Target Audience:
Anyone interested in embodied conversational agents, facial and body animation, gesture coordination.

Prerequisite:
This tutorial does not require any in-depth familiarity with the topics outlined above, a knowledge of 3D computer graphics and animation is sufficient.

Course Length: full-day

Tutorial Intervenants:

Catherine Pelachaud

IUT of Montreuil - University of Paris 8 France

c.pelachaud@iut.univ-paris8.fr