At the object level there are the Domain plan library and the Speech Act
Library.
The Domain plan library contains pre-compiled
recipes which describe typical well formed plans to obtain domain goals.
We chose the University environment as an experimental domain; so, actions
regard borrowing books from the library, taking exams, and so on.
The structure of this library recalls other well known solutions,
so we will not describe it here.
The Speech Act library contains the definition of speech acts: they
are represented, similarly to domain actions, as acts that an agent performs
to try to change the world state (in this case, the
hearer's beliefs). The reason for representing speech acts at the object
level is that, in this way, their planning and execution can be reduced
to that of any other actions, by means of the Agent Modeling plans, and the
treatment of dialogue is unified for any type of interaction.
In particular, we model communication by introducing multiple (object level)
actions, each of which has a different role:
- at the highest level, speech acts are performed to induce the hearer to act
in a certain way Smith-Cohen:96; action ``Get-to-do'' represents this
level and has the effect that the hearer intends to act in a certain
way;
- ``Get-to-do'' is realized by means of one of a set of alternative
illocutionary acts (e.g. a request, an order, or other), which have a weaker
effect:
that it is shared among the interactants that the speaker has the communicative
intention that he intends that the hearer intends to act in a certain way.
``Get-to-do'' can also be realized by means of a complex communicative act,
composed of multiple speech acts, which are related by rhetorical relations
Moore:95,Barboni-Sestero:97b;
- the illocutionary act can be performed, in turn, by means of different surface
speech acts (e.g. a request may be performed as an imperative sentence, a
question, or other). We use direct and indirect speech acts to describe some of
the politeness techniques used in dialogue, following the taxonomy of
politeness strategies in [Brown & Levinson1987];
- the surface acts are all performed by means of a ``Locutionary-act'', which
has as an argument the syntactic and semantic representation of the speech act;
in turn, this is realized by an ``Utterance-act'' which contains a suitable
text string.
As an example, we report the ``Ask-if'' illocutionary act (asking information about the truth value of a condition):
Ask-if