The setting is a small turn-based game of dice where at least one of the player's is substituted by an ECA, the Greta Agent (Bevacqua, Mancini, and Pelachaud (2004); Hartmann, Mancini, and Pelachaud (2002)). The game can only be won by deceiving the other players and by detecting such attempts from the other players. Imagine it is player one's turn. He casts the dice, examines his cast without letting the other players know the result and announces a result. Player two may now believe him and cast the dice herself. In this case she will have to announce a higher result. If player two decides not to believe player one, the actual result is revealed. If the actual result is identical to the announced one, player two has lost this round, otherwise player one.

This demonstration is characterized by a sophisticated combination of multimodal input and output devices that allow for an engaging and immersive interaction with other human and virtual characters. Figure 1 gives an overview of the architecture and the spatial setting of the system. In the current version, the players use a cup of dice with a USB-web cam mounted on top of it. Thus, the game server keeping book of the game progress knows the actual result of the cast. The players announcements of their results are captured by a microphone between them and analyzed by the speaker independent Esmeralda speech recognition system. So far, the possible results and some yes/no variants are recognized. The agent’s behavior is controlled by a Bayesian network that is augmented by a two dimensional emotion model, capturing the aspects of valence and arousal. The players are sitting left and right of the Greta agent which is projected to the wall at the end of the table. Thus, the agent can be said to be sitting with the players around the table. On the left side of the agent a GUI with some game relevant information is displayed. In case of the agent's turn, a short movie of casting dices is shown.

For more information see: http://mm-werkstatt.informatik.uni-augsburg.de/videos/gamblev2_small.avi