Reusable System Architectures - Man-Machine Dialogue: Design and Challenges - page 67

reflective level matching dialogue strategies, has led to the rise in N-tiers
architectures, with new constraints and new system specifications. Rosset
[ROS 08, p. 83] mentions a set of approaches of this type, with the typical
example of double-layer architecture to manage in a simultaneous and
asynchronous manner the short-term behaviors, such as starting to talk, and
the long-term behaviors, such as task and dialogue planning, but also
triple-layer architectures, able to manage various aspects of communication
separately, such as with an ECA. Similar multilayer architecture have also
been used for ages in the MMI field, with interaction management that
separates various logics: the persistence logic that concerns lasting data, the
application logic that concerns task management, the interaction logic that
concerns user action management and the data presentation logic that
concerns real-time display, so as to not present obsolete data or data that has
just been acted on by the user. By rationalizing the design, these architectures
allow us to adapt them to different terminals and different work contexts. The
generalization of this approach to architecture for interaction system in a
broader sense is recent. It allows us to integrate software design models and
man-machine communication models. Thus, Lard [LAR 07] offer a hybrid
approach meant to specify architecture for systems that bring MMI and MMD
together.
4.1.3. Module interaction language
Whatever the chosen architecture, data are exchanged. A format is thus
required to proxy the data in a standardized manner, so that each architecture
module can decode and use what it might need. As for architectures, there
might be as many interaction language proposals as there are systems,
especially in multimodal dialogue, due to the diversity in possible modalities
and the type of content that is used. Denis [DEN 08, p. 93] mentions a few of
them and highlights the Mmil language, which was used in the Miamm,
Ozone, Amigo or even in the Media assessment campaign [DEV 04], in
which he participated. This language takes the shape of a standardized file,
which allows for a representation of the communicating events in an MMD
system, whether we are talking of multimodal external events (speech and
gesture) or internal events (exchanges between modules). The point is to
represent both the events and the content of the messages exchanged. The
representation also takes place at various levels, and it includes, for example,
a representation of the semantic content of an utterance or even its speech act.
The content also remains as close as possible to the utterance without much