The Development Stages of a Dialogue System - Man-Machine Dialogue: Design and Challenges - page 48

that the module requires itself. The semantic representation obtained is
enriched by the reference resolution module that uses a database containing
all the application objects, so the complete representation thus achieved can
be used as a base for the identification of the language act characterizing the
utterance. Thus, we obtain a representation made of an illocutionary value
(order or question, for example) in a semantic context. This representation
enters into the module concerned with dialogue management, which
implements the designer's model: depending on the utterance's illocutionary
value, the system will choose its answer. The generation is simple, uses
patterns, that is sentences with gaps in them, which are predefined according
to the categories matching the different possible illocutionary values. For each
of the modules mentioned, the designer develops his/her own local algorithm
from the available data: its definition of the task, of the vocabulary and of the
matching linguistic phenomena. For text to speech as for automatic
recognition, he/she uses an existing piece of software. All the modules carry
out their tasks on the same computer. The speech recognition and text to
speech modules are demanding when it comes to memory and machine
calculation time. There are thus a few seconds between the end of the user's
utterance and the beginning of the system's response.
The designer can then test his/her system by taking up the progress
imagined for a few typical dialogues on the selected task. Clearly, the results
are not good. He/she has to take up the settings, or even the development, of
each module again. To refine the understanding abilities, he/she rewrites part
of the resources. After a while, he/she starts the tests again, playing the role of
the user like he/she did during the development phase. No metric assessment
is used. Intuitively, and also because it would be a shame to throw the system
out after so much effort (especially since nothing can be reused for a different
system), the results seem to have improved. The designer realizes, however,
that all the words defined in the lexicon have not been used: on the contrary,
the system can only understand and process a small subset of the lexicon. No
matter, it is now possible to dialogue with the system. Even if the resulting
communication is rather simplistic, it has the advantage to be possible.
3.1.2. A scenario matching the 2000s
In the 2000s, various designers came together within the framework of an
international project or a perennial laboratory activity sustained by constant