Man–Machine Dialogue Fields - Man-Machine Dialogue: Design and Challenges - page 33

point is that with this, the system can understand and provide compensation
for the user's biased or incorrect perception.
It is the same thing with attention phenomena. The cognitive sciences have
clarified the different forms and functions of attention [GAO 06, p. 137]:
attention retention (reacting to new things and maintaining awareness);
attention selection (filtering certain types of information); attention allocation
(managing simultaneous activities); attention control (controlling the progress
of actions, i.e. preparing, alternating, supervising). Providing an MMD with
such abilities and limits only makes sense if you are trying to reproduce
human fallibility. If on the contrary, you are trying to design an
understanding, performing system able to use any clue that allows it to solve
the task for which it was designed, there is nothing to prevent all the recorded
information to be processed, without selecting or allocating the system's
attention. However, in this case as well, the human attention characterization
is a result that the system can use when it generates a message for the user.
Memorizing and forgetting give rise to different questions. In the models
stemming from cognitive sciences, various types of memory are
distinguished. At the level of immediate perception and the preliminary
mental representations, we talk of very short-term or short-term memory.
This is the level at which the memory span comes into play, a threshold
limiting the number of elements stored in this memory, with the consequence
of preventing a human from rendering a greater number of elements. At the
level of the first mental activities, we talk of long-term memory and, for
example, distinguish between semantic memory, which stores general,
verbalizable knowledge, from episodic memory, which store specific events
that happened at a given time and left a mark in the user's mind. In MMD,
taking into account that memory span is a strategy for knowing the user's
limits so that he is not overwhelmed with information, for example during the
message generation phase. It is thus applied in the same manner as previously
mentioned. However, the nature of semantic and episodic memories can also
provide ideas during the system's design itself: it is, for example, a way for
remembering to save any specific information during a dialogue, such as past
events; it is also a way for separating the different sources of knowledge in
the system's architecture, and to specify the encoding processes and recorded
information retrieval. More than that, the notion of forgetting can be
interesting: after a certain amount of time or a certain amount of speech turns,
we can consider that a piece of information that was transmitted and not