A Few Dialogue Strategies - Man-Machine Dialogue: Design and Challenges - page 138

acts, that the identification of the user's mental states contributes reasoning
possibilities and the management of a correctly structured dialogue history is
also indispensable. As an example, Vilnat [VIL 05] suggests the notion of a
history page that contains the speaker's identifier for each intervention, the
semantic and pragmatic representations, the topic concerned, the goal
concerned, the state of the dialogue's structure, the state of the interaction
variables, the state of the plan being developed, etc. As we can see, the history
page is a complex, multiform and multifunctional structure.
If we take up the three main speech acts, we can consider the following
processes and structures:
- “saying that”: the speaker expresses an assertion in order to let the system
know something. The system updates its database which is part of the common
ground and can be referred to as CG. Indeed, by providing the system with
information, the speaker helps to make this information mutually known and
manifest (at least after taking the grounding process into account);
- “telling to”: the speaker gives an order so that the system does something.
It updates a list of actions to be carried out, which we can refer to as a to do list
(TDL). It is a sort of pile (or heap) cataloguing the things to do as the dialogue
progresses and taking an item out as soon as the matching action has been
carried out. Managing this type of list allows the system to know what it still
has to do without it being linked to the current utterance's processing but on
the contrary, making it possible to carry out an action several speaking turns
after the user's request;
- “asking”: the speaker expresses a question in order to learn something
from the system. It updates the list of questions which it has to answer, which
we can call questions under discussion or questions under debate (QUD), that
is a similar structure to the TDL, in charge of cataloguing and managing the
questions asked during the dialogue.
Depending on the approaches, only one of the three structures can be
updated during a speech act's processing or, on the contrary, multiple updates
are allowed. In his successive models, Ginzburg [GIN 12] detailed the
operations of substructures and suggested complementary structures, for
example Latest Move (LM), Shared Ground (SG), F ACTS and P ENDING
(transient structure). Other approaches highlight a structure cataloging the
speaker's commitments, Commitment Store (CS), a structure specific to