A Few Dialogue Strategies - Man-Machine Dialogue: Design and Challenges - page 135

the semantic and pragmatic analyses have been carried out, that is at the level
of the role of the utterance in the dialogue's progression and, for example,
in satisfying the task. Among these elements we can thus find hypotheses on
the user's beliefs or desires, which can help put the semantic content of his/her
utterance into perspective. We also find all kinds of information using what has
already been said during the dialogue, and thus calling upon the history. This
is the second aspect of dialogue context modeling: formalizing and storing in
a data structure the contextual interpretation results as well as highlighting in
this history a structure describing the dialogue's progression, for example the
path followed to solve the task.
A notion that has emerged and generated many proposals in theoretical
work on dialogue is that of common ground, that is the information set shared
by the speaker and hearer, either because it mutually manifests, for example
because it has been verbalized, or because it can be deduced from what has
been said: for example, when the system answers “to Paris. When would you
like to leave?”, it is obvious for both the speaker and the hearer that the user
wants to go to Paris and the system is aware of this desire. The common
ground becomes what the speaker and hearer build as the dialogue progresses.
Therefore language is seen as a joint action [JUR 09]. More precisely, we can
distinguish between a communal common ground, where the shared
knowledge goes beyond the interaction between the speaker and the hearer,
and a more personal common ground in which the knowledge is only good
for the speaker and the hearer [DEN 08, p. 45]. To be part of the common
ground, a speaker's utterance must be grounded by the hearer. We thus
distinguish between the grounding process, a process during which the
speaker and hearer update the common ground, and the grounding criterion,a
criterion which the speaker and hearer are aiming for, with a desire for a joint
belief of understanding. Several grounding models have been suggested. The
first models, which do not draw any of these distinctions, automatically
update the common ground at each utterance. As a reaction, Clark and
Schaefer [CLA 89] suggest the discourse contributions model, in which the
grounding can only happen when the grounding criterion is reached. Other
proposals try to go down this path to reach implementable models, especially
the grounding acts model with its nine levels of grounding, from “unknown”
and “not understood” to “accepted” [TRA 92], then the weak belief model
which provides an explicit modeling of understanding beliefs necessary for
grounding: it considers that the speakers and hearers carry out hypotheses on