The Development Stages of a Dialogue System - Man-Machine Dialogue: Design and Challenges - page 61

- Going from limited and highly reliable resources and data to real data, of
great size with potential errors, noise and non-homogeneity.
- Going from a system's occasional execution for a relatively short period
of time each time to a continuous operational mode involving a minimum of
rebooting.
- Going from a user mode allowing a certain margin of error and
dysfunction to a use that supports no dysfunction tolerance and has a very
limited tolerance to mistakes.
- Going from a user mode involving a user both framed and willing (often
an expert) to a mode involving a demanding and sometimes malicious user,
always ready to play with the system or even to try and break it by any means
possible. There is nothing more distressing for the designer than to watch such
a user, who is referred to as the final user, torturing his/her system.
The notion of re-implementation comes into play after the assessment. It
consists of modifying the system's code so that it can process the few problems
it has identified as the most common or the most problematic. The goal is to
minimize this phase of code updation, and do anything so that at the beginning
of design, it has to be the easiest and fastest activity in the world. Rieser and
Lemon [RIE 11, p. 16] underline that today, there is no rational method in
MMD to turn assessment results into code. This is however the case in other
computer science fields, for example computer graphics and image synthesis
(the notion of relighting after the full calculation of a rendering), and this is
again an essential methodological challenge for the years to come.
3.3. Conclusion
When we start developing a dialogue system, we do not start to program
directly and let the issues crop up as we go. Instead, anticipation is crucial
and many stages of work are carried out before any kind of computational
development: defining a task that matches the goals that the system must
fulfill, observing and analyzing the human dialogues focusing on such a task,
determining the linguistic and interaction phenomena that the system must
process, carrying out experiments, specifying the system's software
architecture, etc. This chapter provides examples of development scenarios
including these stages before being implemented in development, testing and
assessment.