A Few Dialogue Strategies - Man-Machine Dialogue: Design and Challenges - page 139

salient utterance, S AL -U TT (Salient Utterances), or even a subset of the QUD
structure in charge of a specific type of question, or I SSUES , following a
distinction between various types of question depending on their function in
the dialogue [DEN 08].
8.2.3. Dialogue management and multimodality management
The previous example had a dialogue management initiative (see
section 8.2.1): when the system decides to answer “here are the possible
itineraries” and displays a set of alternatives on the visual scene, it makes a
choice about what it is about to say and display. The display of possible
itineraries and the direction to Paris fall under the scope of an act of a “saying
that” type, except it is “saying visually” instead of a “saying”. The utterance
of “here are the possible itineraries” falls under the scope of a “saying that”.
The system could have made other choices, such as asking a question like
“would an itinerary with a change in Meudon be acceptable?” which would
advance the task and test one of the alternatives. The utterance would then be
characterized by the speech act “asking”. The speech act choice in reaction to
the user act is thus part of dialogue management. In the multimodal context
that we are studying here, the chosen act is a multimodal composite act: the
system wants to indicate various possible choices to the user and does it in a
certain way that strongly involves the functionalities of multimedia
information presentation.
Dialogue management has other aspects that our example does not give,
for example clarifying the conditions that allow the system to abandon its
goals, decide on a path aiming to satisfy a specific goal and in general explore
the interaction between goals, beliefs and intentions. This is what Cohen and
Levesque [COH 90] show with the example of a robot that states it is going to
bring something, does not, and then explains that it found something else to
do, and this example illustrates rational action theory. Moreover, dialogue
management also includes the ability to manage incomprehension (see section
8.1.5 with the same type of preoccupations for ambiguity management). The
system can choose to solve the incomprehension without the user (we then
refer to internal robustness) or to solve the incomprehension with the user,
that is by starting a clarification sub-dialogue (external robustness). As Denis
[DEN 08, p. 35] shows us, the two approaches are complementary: we cannot
make do with a very good internal robustness and a weak external robustness,
as the dialogue also has a function of talking about what is not going well,