7.2 Possible Refinements of the RM

There are a number of possible refinements and extensions to the current RM. Two of them are sketched below. Further ones are discussed in the contributions by (... to be added).

7.2.1 Moving towards a Formal Specification

A further phase in the development of a complete formal reference model is to express in a semi-formal notation the functionalities of each black box component of the system. As an example, let us consider the Goal Refinement component in the Content Layer. Below we sketch a semi-formal specification of parts of its interactions (in the style of CSP [17]).

The specification expresses the following matter of subject: The Goal Refinement component non-deterministically receives goals from the Control Layer's Goal Selection component (line 1). Received goals are decomposed (2) and stored in the context knowledge in the Context Expert (3).

If the Content Selection component is ready to proceed with a new goal (4), and so is the Media Allocation component(5) and there is such a new goal reported from the Context Expert (6), then the goal will be actually forwarded to the Content Selection (7) and Media Allocation (8) components.

Goal Refinement also receives notifications (9), from the Linearization component. Notifications are analyzed: if they report success (10) then they are elaborated (11) and eventually propagated upwards to the Control Layer (13). If any problem occurred then a repair strategy should be determined (14) which is pushed into the Context Expert (16). Finally, if no repair strategy is applicable then an elaborated version of the notification (17) will be forwarded to the Control Layer (18).

7.2.2 Moving towards Multimedia Dialogue Systems

As already emphasized in the introduction, the current RM is rather limited in its abilities to handle user interactions. To be more precise, the RM only covers interactions of the following types:

• direct triggering of presentation goals;  i.e., there is a direct mapping from a user input to a presentation goal. For example, the system may provide menus from which the user can select the presentation goals which the system shall accomplish.
• presentation unrolling;  i.e., the system designs a branching presentation structure and leaves it to the user how this structure is traversed. Examples of such presentations are hypermedia documents with statically linked parts, but also presentations in which the target of a link is only generated on demand.

A full-fledged multimedia dialogue system would allow the user to express her/himself through the use of different modalities and media. Such interactions are obviously beyond the scope of the current RM. However, if one considers multimedia analysis being somehow reverse to multimedia generation (see also [18]), one can obtain a rough task decomposition just by turning the RM upside down, and replacing the generation subtasks by analysis subtasks. For example, at the counterpart to the Presentation Display Layer one would find media input devices and a component which tracks the spatial/temporal relations of user input in different modalities. Medium and modality specific analyzers will be found at the layers which correspond to Realization and Design. Instead of Media Allocation, one would have a component for fusing media specific communicative acts in order to bring about the underlying semantics and eventually the user's communication goals. Future work will show to which extent such a reversability approach will lead to useful results.