An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 13

unlike GUS or the example we gave in the introduction, it involves various
modes of transportation and thus manages the connections between these
modes, the planning issues, the optimizations (journey length calculations),
the potential conflicts, etc.
In France, systems and publications are multiplying [BIL 92, GUY 93,
DUE 94, LUZ 95, SAB 97, GRI 00] and we will remember as an example the
Dialors system by D. Luzzati, which focuses once more on train ticket
reservation. The methodology starts here again with an in-depth corpus study,
in this case a corpus coming from Société Nationale des Chemins de Fer
Français (National Society of French Railways SNCF) recordings, a corpus
that has also been the focus of various publications. The Dialors system has
an analyzer called Alors whose function is to turn utterances into an internal
representation to the system, and a dialogue manager, Dialogue, who decides,
depending on the representation, on the action to be carried out: request
clarification, answer the user's query after consulting the train timetable
database. The second component has the role of implementing the dialogue
model suggested by the author, a model that distinguishes the governing
dialogue, i.e. the main dialogue reflecting the task's progression, from other
potential incendental dialogues, i.e. the clarification requests and other
transient sub-dialogues, which do not influence the task's progression but
allow the interlocutors to understand each other. This dialogue structure
allows the system to carry out fine analysis and also to assess in real time the
task's progression, without requiring the implementation of a more complex
model such as the hierarchical model of the school of Geneva [ROU 85],
mentioned earlier as an approach to discourse analysis.
1.1.3. Current systems: multiplicity of fields and techniques
Our overview started in the 1950s and now reaches the 2000s. It is harder
to use hindsight on this period that includes the current systems, especially
since the work has multiplied and the number of techniques has increased.
In general, beyond the improvement of all the models of the 1990s [JUR 09,
p. 892], here is what appeared in the 2000s:
- the application of computer techniques of machine learning to MMD to
relocate part of the different settings onto the big corpus processing or onto an
improvement of the performances as the system is used [RIE 11];
- the tremendous efforts of standardization: W3C, ISO, TEI, DAMSL, etc.;