Abstract: |
With the growth in technology, many applications supporting more transparent and flexible human-computer interactions have emerged. This has resulted in an increasing need for more powerful communication protocols, especially when several media are involved. Multimedia multimodal applications are systems combining two or more natural input modes, such as speech, touch, manual gestures, lip movements, etc. Thus, a comprehensive command or a meta-message is generated by the system and sent to multimedia output devices. Each application is based on a software architecture combining modalities to match and elaborate on the relevant multimodal information. Such applications remain strictly based on previous results, however, and there is limited synergy among parallel ongoing efforts. Today, for example, there is no agreement on the software architectures that support a dialog implementation, independently of the application type. The main objective of this thesis is twofold.
First, we propose new architectural paradigms for analyzing and extracting the collective and recurrent properties implicitly used in such dialogs. These paradigms use the agent architecture concept to achieve their functionalities and unify them into software architecture modeled, specified and refined by stochastic timed colored Petri nets.
Second, we propose a new methodology to show the ways in which agents can be introduced at the architectural level and how such agents improve some quality attributes by adapting the initial architecture via multimodal scenarios. Examples of multimodal prototypes are sketched to illustrate the proposals. |
