Man–Machine Dialogue Fields - Man-Machine Dialogue: Design and Challenges - page 44

For these questions, MMD systems implement local procedures, which depend
on the task and operate without providing any kind of general solution.
2.3. Computer aspects
The computer aspects involved in designing an MMD system are
numerous, as the history of the field has shown, notably the processing of an
audio signal, the processing of recorded gestures through a specific device,
face tracking and more generally artificial vision, NLP, knowledge
representation, inference engines, machine learning, text to speech synthesis,
multimedia information presentation, or even the annotation and statistic
study of corpus. All these goals require theoretical proficiency, be it
linguistic, psychological, ergonomic, logic, mathematical, and involve
specific data structures and algorithms. In Chapter 4, we will explore a
different field of computer science, that is software architecture and
model-driven engineering. We can see that the panel of technical skills
necessary to implement an exhaustive system is extremely vast. As in the
previous section, here we will very quickly present a few landmarks on two
aspects chosen randomly: digital resources and plastic MMI.
2.3.1. Data structures and digital resources
The era of the 14 Kb database of the PARRY system is long gone. Today,
we seek to maximize the amount of data recorded in the system, to provide
the maximum amount of material to analysis algorithms. The achievements
publicized by IBM, such as Watson, are proof of this; the data and brute
calculation strength allow us to achieve operational systems. This, at least, is
one of the ways to achieve it, as the reflection on data is still essential. In NLP
and, in general, with the rise of the semantic Web, efforts have been made for
years to improve the quality of the data. This goes from using standardized
languages to adding metadata to data that describe the content in a sober and
universal manner, or to ontologies, set structures of terms or concepts
common to a field or a task. In MMD, we kept a closed operation for a long
time, with data internal to the systems and not connected to the Web. On the
contrary to resources available on the Web, which are in constant evolution
and can reach unreasonable sizes, these partitioned data have the advantage of
being very quickly accessible, in real time, an aspect imposed by spontaneous
MMD. The data, however, require being first specified, and one of the