An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 22

the issue also includes that of the graphical presentation of information (such
as the Immps mentioned earlier). It can happen that the MMI itself generates
earcons. The earcon generation and the speech generation must then be
carried out in a relevant manner, i.e. without superimposing them, which
might hinder the user's perception.
1.3. The current challenges
Can a machine think? The Turing test and Grail-style quest for the talking
machine is still as fascinating as it ever was, but the current limits to the
MMD systems mean that the question is not set in these terms any more. In a
more pragmatic approach, the questions are set in terms of limits in the
abilities of the machine to model and naturally process natural languages, to
represent and reason on logical representations and to process and integrate
various and varied signals. Commercialized versions prove this everyday: an
MMD system only works well within a limited application framework, that is
in a sufficiently limited framework for the maximum interaction possibilities
to have been imagined upstream, during the design phase. Contrary to what
attempts such as ELIZA had us believe, nothing is magical, and no matter
what technologies are implemented, whether or not they are symbolic,
statistical, or involving machine learning or not. Everything must be
anticipated, and this represents an amount of work proportional to the
considered abilities for the system.
The main challenge of MMD remains, as it was in Pierrel's time [PIE 87],
the multidisciplinary design of comprehensive systems that allow the user to
express himself/herself spontaneously as he/she would with a human
interlocutor, and this for a variety of applications, so as to offer systems that
are accessible to anyone, in all everyday situations. More precisely, we will
develop four sets of challenges: theoretical challenges, challenges concerning
the span of expected abilities in a system, technical challenges concerning
system design and technical challenges trying to help system development.
1.3.1. Adapting and integrating existing theories
According to Cole [COL 98, p. 191], recent strides have not included the
development of new theories but focused on the extension and integration of
existing theories. Thus, we can find many hybrid approaches that use the
expressive power of more than one existing theory. This observation that we