Man–Machine Dialogue Fields - Man-Machine Dialogue: Design and Challenges - page 29

mentioned in the previous chapter, especially those that arose from the speech
act theory [AUS 62, SEA 69] and the dialogue hierarchical structure
[ROU 85]. In general, the approach tries to formalize the interpretation of
natural language and the management of the dialogue is as follows: it starts
with the linguistic study of (written and oral) communication phenomena,
their methodical analysis, classification and characterization, i.e. the
identification of their relevant aspects. This phase requires general linguistics
and sociology, which provides additional explanations to the functioning of
speech turns and polite forms of address. Then comes the modeling phase,
which consists of identifying the rules that allow us to take into account the
maximum amount of phenomena. This phase requires formal linguistics,
NLP, and computer science as soon as we reach formalization and
implementation. It comes with hypotheses, to determine rules left unclear by
the theories, and then borrows experimental protocols from psycholinguistics
to test these hypotheses. Finally, the dialogue management and message
generation to the user involve decision making for which psychology and
cognitive sciences are crucial. For example, it is by taking into account
concepts of cognitive load and work memory that the system can adapt itself
to the human user's abilities.
This chapter, which we can approach as a list of the fields we should know
before starting in the MMD field, draws a summary of the cognitive aspects
(section 2.1), linguistic aspects (section 2.2) and computer science aspects
(section 2.3), the latter becoming numerous as the system processing abilities
get bigger.
2.1. Cognitive aspects
An MMD system is an example of a cognitive system, i.e. a system
characterized by the facts that it processes outside data, that it has knowledge
and that its behavior is based both on this knowledge and the processed data.
Depending on the way the cognitive system is modeled, we can follow the
structural cognitive science path, which highlights structures and mechanisms
of structure operation, or the path of computational cognitive science, which
highlights the processing of an information flow and tends to assimilate
humans, as cognitive systems, with artificial cognitive systems.
Cognitive sciences cover the fields that study natural or artificial cognitive
systems. In fact, they are also the fields that focus on language as a human