CrossTalk       CrossTalk     Publications     Architecture     Video     Contact     Development     Internal

Architecture

The CrossTalk architecture depicted below only deals with the online processing of presentation. After having assembled a complete presentation script in the plan-based JAM language, using our authoring tool called SceneMaker, the system can process the presentation in the PresentationProcessor module. This central module has access to the dialogue memory, the emotion engine and the output renderer. The underlying communication platform, the InfoRouter, allows modules to subscribe to message types and post messages in the multi-blackboard communication paradigm. The DialogueMemory component keeps track of the ongoing presentation and interactions and yields access to the discourse history. Also, situative context like current time and location is made available through this module. The EmotionEngine contains a powerful reasoning engine, implementing the OCC model of emotions, a cognitive paradigm that allows mapping so-called emotion-eliciting conditions to an intensity vector of basic emotions. The plan-based generation engine contained in the PresentationProcessor generates multimodal action scripts that control textual, gestural and other screen output. This action script is interpreted at runtime by the output renderer to produce commands for the specific renderers for screens, input devices and agent animation players.