Reusable System Architectures - Man-Machine Dialogue: Design and Challenges - page 63

conceptual run-time architecture is probably the most important characteristic
of a system. It allows us to describe it and to show what its characteristics are.
The conceptual run-time architecture is a kind of leaflet that highlights the
system's innovative aspects and that proves it is not a quick improvement of
an old existing system. There are multiple consequences that lead to two other
observations. Second observation: despite the wish to reuse existing
resources, each system is defined by its own run-time architecture. It is
directly linked to processing, understanding and generation abilities, and
since all systems do not run with the same inputs and output, it is thus logical
that the architectures are different. However, the current techniques allow us
much more flexibility than before as to the operation of partially implemented
architectures, and we should be able to rely on reference architectures more
easily. It remains that the proposal of a new system is often accompanied by
new architecture, with the leaflet and innovative effect mentioned earlier.
Third observation: whereas the specification of an architecture should help
collaboration between researchers, it is, on the contrary, sometimes a source
of problems. Each researcher has his/her own suggestion, and conciliating
approaches can be a long and delicate process. More than that, it happens that
some people have a tendency to try and include other people's proposals as
modules in their own architecture. What comes from these issues is that there
is a lack of generic, reliable and trustworthy reference architecture that can be
applied to any system. The rise in design-time architectures, if it is confirmed,
might help give new solutions to this problem.
This is what we will discuss in this chapter, with sections 4.1 and 4.2 on
run-time architectures and design-time architectures, respectively.
4.1. Run-time architectures
4.1.1. A list of modules and resources
López-Cózar Delgado and Araki [LÓP 05, p. 5] show us a very complete
run-time architecture, which notably integrates modules to process specific
modalities such as lip-reading or tracking the user's face. In general, all the
topics mentioned in the introduction present architectures. These are generally
diagrams made up of boxes (modules) that are labeled (processes) with arrows
between certain boxes (processing sequence) that can themselves be labeled
(type of data exchanged).