Reusable System Architectures - Man-Machine Dialogue: Design and Challenges - page 71

the man-machine interaction in the original multilayer architecture. This
layer is implemented as interaction middleware and provides generic services
for man-machine interaction, as well as an abstraction of application
specificities and uses of contexts. This proposal focuses on the technical
aspects with regard to the general architecture and practically does not
mention the NLP issues and natural dialogue in natural language issues.
However, this is a path that is a challenge to help the MMD system
development process.
4.2.3. Challenges
As we have seen, the paths taken toward running and publishing
design-time architectures are still in their infancy. They allow us to draw up a
list of the challenges for the years to come more than to draw a conclusion on
the advances that have actually taken place. Indeed, the design of many
current systems bypasses the design-time aspect, the specification of a
run-time architecture rallying most of the efforts.
In the previous sections, we saw that one of the first stakes was the
establishment and broadcasting of the development environments that were
best adapted to MMD, that is to the NLP aspects in all their complexity, right
up to natural dialogue management techniques. A second challenge consisted
of applying the MDE approach to the MMD, taking into account once more
the diversity of aspects involved, and thus implementing a large variety of
models, which tend to echo the list mentioned in section 3.2.5, by adding all
the technical aspects, and notably the system adaptation aspects (plastic user
interfaces). A third challenge is the specification of an interaction middleware
that has to be more elaborate than the challenge presented in [LAR 07], with
once again taking into account NLP aspects in all their variety and
complexity. This challenge can actually be multiplied if we consider that such
a piece of middleware would be useful for each of the domains of NLP, for
ECA or even for AI with a set of services devoted to different types of logic
reasoning.
Finally, one challenge that is often put forward in AI is the design of
systems able to assess and modify their own algorithms. With the derivation
principles of models and model-driven architectures, this old dream might
become reality. Indeed, an MMD system obtained by deriving models may,