Key Action :
Research Area :
Action line :

ABSTRACT : MagiCster is concerned with the development of believable conversational interface agents which make use of gaze, facial expression, gesture and body posture as well as speech in a synchronised fashion. The project also intends to evaluate the use of conversational agents in laboratory conditions to determine which aspects os the embodied agent are important for what types of human-computer interactions. Finally, the project aims to develop and document the agent architecture and components to enable other research and devlopment teams to prototype and evaluate new versions of the agent interface in new domains and for novel tasks.  MagiCster is concentrating on the following key technologies:

• believable animation: developing an agent which displays natural or believable movement
• believable conversation:
developing an agent which can perform conversational functions such as repair, feedback or turn-taking in a believable way using verbal and non-verbal signals
• believable information delivery: constructing an agent which can deliver information in a contentful and contextually appropriate way
• evaluation: evaluating the use of believable agents in information delivery

OBJECTIVES : People are spending more and more time interacting with computers, which raises a number of questions about the design of communication interfaces. Speech is often seen as the most natural, but this is only part of the answer as there is evidence that it is the ability to engage in conversation, in particular face-to-face conversation, which allows for the most natural interactions. Conversation involves more than speech. A wide variety of signals such as gaze (making eye contact, looking down, looking at an object ... ), gesture (pointing, indicating the size of an object ... ), or body posture (turning away, leaning forwards ... ) are used in natural interactions. Building speech-based interfaces which do not allow for these other signals will not allow for high-bandwidth communication and will inevitably lead to interactions which break down easily. The main goal of the project is thus the design of properly conversational interfaces, in particular those which make use of non-linguistic information. We aim to allow users to interact with an agent-like character which has as visual and other properties (a body, a face, lip movement, moving limbs, and so on). We call such representations embodied agents or characters.

DEVELOPMENTS : The work will concentrate on developing the following key technologies:

• Believable Animation. The development of an agent which displays natural or believable movement. This includes modeling not only voluntary movement (such as pointing) and involuntary movement (such as blinking) but also movements which correlate with communicative acts (limb gestures, body posture, facial expressions, and so on).
• Believable Conversation. The investigation of an agent which can perform conversational functions such as repair, feedback, or turn-taking in a believable way, using verbal and non-verbal signals. This work will build on empirical and descriptive work on person-to-person dialogue, as well as on computational work on end-to-end spoken dialogue systems.
• Believable Information Delivery. The development of an agent which can deliver information in a contenful and contextually appropriate way, building on earlier work on the generation of spoken and written messages and integrating this with work on animation. To increase believability, the project aims to pay particular attention to the generation of messages in a linguistic style which projects the right kind of `personality'.
• Evaluation. The evaluation of the use of believable agents in information delivery. Throughout the project various aspects of believability will be developed and instantiated in different combinations in subsequent prototypes. This will allow us to evaluate the effects of different types of believable agents in different contexts. We also want to work on the development of an evaluation methodology for MagiCster-style technology.