1. Problem can be separated into input vs. output. SMIL addresses output, but no solution exists for input
2. On the output side, we need to synchronize rendering across modalities. One view needs to know when the other view is done rendering
3. Latency (esp. with distributed architecture)
4. Synchronization information needs to be authored
5. Multi-modal input events need to be time-stamped, which requires clock synchronization. Time-stamping of output (prompts) also may be needed. (DOM2 is not a real-time model)
6. We don't know how to get access to the synchronization events
7. Ergonomics might require "adjustment" of synchronization
1. A model for synchronization is needed, and a corresponding markup
2. Time-stamping for synchronization
3. One proposal: Model/view/controller
a. Build from DOM level 2 for VoiceXML, WML
4. Confirmation UI may need to be blocking. If so the DOM2 event model will need to be modified, for example defining prioritization of input events
5. Blocking/non-blocking of rendering (at the User Agent) should be an implementation choice, not an aspect of the model
6. Improvement on output synchronization also needed, e.g. for talking heads
1. Specification of dialog must be independent of the modality
2. Combination of multi-modal inputs
3. Conflict resolution
4. Dialog management
5. History, context
1. Single authoring is probably the easiest way to build a model, e.g. model/view/controller
2. Event interface with modalities needs to be standardized (model itself may not need to be standardized)