An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 25

unease and disturb the user. The field of robotics or that of
computer-generated images use the term of uncanny valley to describe this
type of phenomenon. The issue is that we try and get closer to the human (to
reality for computer-generated images) but there is still a small gap between
what is achieved and what is aimed for. And this gap, as minute as it might
be, is enough to be perceived, and to irritate. To counter this, some designers
make the gap visible and forego the goal of getting close to the human. So
some mechanical toys that look like animals do not have any fur. In MMD,
for example, the Web service Ananova takes on the appearance of a gorgeous
young lady ... with green hair [HAR 05, p. 341].
Finally,akeychallengefortheabilitiesofanMMDsystemisitsrobustness,
that is its ability to manage its own shortfalls, at a linguistic analysis level for
example, its own deficiencies and errors, and its ability to always bounce back,
to help the dialogue progress no matter what the cost, by using the task to
be solved, or not. This implies the design of modules able to operate with
incomplete entries and have strategies to manage problems. This also implies
ability to predict, from the first stages of design, tests and settings with real
data, real conditions, rather than laboratory-controlled conditions.
1.3.3. Rationalizing the design
At the design level, there are multiple methodological and technical
challenges. Once the list of understanding and generation abilities is
determined, they have to be instanced and organized into modules,
components or agents in an architecture, and the interaction languages
between these elements, the evaluation methods and construction methods of
necessary resources of integration have to be specified. The main challenge
here is the rationalization of the architectures' engineering (see Chapter 4),
and in general the rationalization of production flows, as in any professional
technical field. Thus, Harris [HAR 04] focuses, Chapter 9, on a very precise
description of a design team, with the different professions involved: a
dialogue team leader; an interaction architect; a lexicographer in charge of the
aspects linked to corpus; a scriptwriter in charge of anticipating the expected
types of dialogues, but also the definition of the system's personality and its
possible reactions; a quality engineer, without forgetting the ergonomics
experts, technical experts as well as an expert in the field covered by the task.
The task, and more generally the context of the dialogue, can require
integration into another field of research. A first example is robotics, in which