Reusable System Architectures - Man-Machine Dialogue: Design and Challenges - page 68

interpretation. The semantic content is thus very close to the utterance's
structure, and does not integrate any representation of the implicit or any
formalization in a logical form: it is the choice that has been made, the
enrichments are managed by the modules in charge and not at the global
architectural level.
One of the aspects of Mmil has also been the focus of various works
linked to run-time architecture materialization. They are the progressive
overlapping of man-machine interaction and module interaction, which is
purely computational. From the point when the man-machine communication
is formalized as commands and queries covering semantic content, there is
nothing to prevent us from doing the same for communication between two
modules in an architecture: a module that needs information to finish an
analysis may query another module just like a module that has just finished an
analysis may communicate it as an assertion. Even if this is not an essential
challenge, since the complexity of man-machine communication is much
greater, it is a way of applying pragmatics research to the design of a
consistent interaction language.
4.2. Design-time architectures
Generic MMD system research, that is partly set and reusable, can go
through the specification of a generic run-time architecture. In this case, a
specific amount of attention is given to determine the modules and
sub-modules, the interaction language and the task's setting procedures. The
latter is specific to the system, and all that concerns the task is managed in an
independent manner. The task objects are grouped in a database that becomes
one of the domain model's settings, the task's specific lexicon becomes a
setting of the lexical module, and so on. If the task allows utterances with a
specific syntax, with, for example, the disappearance of certain prepositions
and certain articles, as it can be seen in specialized languages, then these
specific syntactic structures also become one of the parameters of the
syntactic module. We then separate all that is task specific (settings) from all
that is common to the whole system (knowledge and processes).
Another way of solving the issue of genericity is to put efforts into
designing an MMD system development environment. A system is not run
autonomously, but becomes the product of a kind of system factory. This