A Few Dialogue Strategies - Man-Machine Dialogue: Design and Challenges - page 136

the understanding of their partner and that the confirmation of these
hypotheses allow them to reach the grounding criterion. A reinforcement
mechanism then helps transform a mutual weak belief into a mutual belief;
see [DEN 08]. Finally, a more recent tendency consists of integrating into
grounding models numerical aspects with, for example, the calculation of a
score which allows the system to characterize the point of grounding a piece
of information. As is often the case, this numerical approach complements
symbolic approaches.
If we now take up our favorite example and the distinction of three main
speech acts from relevance theory, the reception by the system of the
utterance “I would like to go to Paris” leads to the set of following reasonings:
analysis of the utterance's meaning which leads to the identification of the
expression of intent. The system thus remembers this intent and this will
allow it to plan its next actions. The system's roles are to inform the user; here
on the ways to reach Paris, and sell him/her train tickets. It is highly probable
that the utterance of “I would like to go to Paris” by the user is an expression
of a desire that the system help him/her reach Paris. Satisfying this desire
means identifying, among the set of possible ways to reach Paris, that which
will best satisfy the user. For now, the system does not know his/her
preferences, however, it can always sort the itineraries from its database to
suggest the most relevant journeys.
At this point, the system thus decides to suggest something to the user and
includes several alternatives so as to give him/her a choice. The system also
decides to manifest an acknowledgment and use multimodality, that is to
display several itineraries leading to Paris on the screen (and highlight the
Paris train stations involved) at the same time as it utters a relatively short
sentence, “here are the possible itineraries”, a sentence that does not include
the information on the trains which is too long to verbalize. Thus, the
utterance's grounding by the user is carried out by highlighting the Paris train
stations, and the acknowledgment of the user's mental states is shown in the
oral utterance. We can consider this answer to be cooperative: it is neither too
short nor too long, it is relevant given the user's utterance, involves the
recognition of an intention and an indirect speech act and it globally satisfies
the theoretical criteria described in the previous sections. As well as
generating this answer, the system stores in its modeling of the dialogue
history the identified mental states as well as this first materialization of a plan
consisting of identifying a way to get to Paris to sell the matching train ticket.