Discussion groups examined three issues: data acquisition and recognition, modeling, and control. From these emerged a broad consensus on several points. First, it became apparent that many data bases already existed which were of wide interest. Efforts were instigated to make these available on the InterNet. Second, there was wide use of the Facial Action Coding System (FACS) for controlling emotional expression; this in spite of the fact that, more generally, there was such diversity in modeling and control that it would be challenging to present a unified framework. Third, tongue and timing were important to both analysis and synthesis, and much more data was needed on these features. Fourth, the level of detail available in FACS was not adequate to represent lips, tongue, and timing. It was clear that some augmented notation was needed, as many participants had already gone to the use of muscle level description or independent augmentation. Speech requires more detailed lip shapes than FACS provides, and even emotion requires more precise timing. Finally, there was consensus that standards could be helpful, both to facilitate data exchange and to support validation efforts. This last point is particularly important, as validation was seen as necessary and desirable in the work of almost everyone present.
In light of the number and diversity of interests represented at the workshop, it is perhaps remarkable that there was any consensus at all. Yet a common feeling expressed at the end was that this bringing together of colleagues, many previously unacquainted, had been enjoyable, educational, and valuable.