Using Misunderstanding and Discussion in Dialog as a Knowledge Acquisition or Enhancement Procecss - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 70

interlocutor's answer. This is why an opportunistic model of linguistic actions is better than a
planning model.
Since tutoring learning seems to be a good framework for knowledge transfer, we have chosen a
particular tutoring action, the socratic dialog . It is a simple conversation structure in which transfer as
well as checking is done in the way of questions on behalf of the teaching agent. For instance, a socratic
dialog might begin with: What do you know about X? and then answer, and then Why is Y related to
X? where Y was present in the answer, etc.
Some f eatures of Tutored Learning Situations
Dialog Peculiarities. Even reduced to socratic dialog, a tutored learning situation implies a finalized
dialog (aiming at carrying out a task) as well as secondary exchanges (precision, explanation, confirma-
tion and reformulation requests can take place to validate a question or an answer). Therefore, speech
acts appear as crucial elements in the interaction process. We have chosen to assign functional roles
(FR) to speech acts since this method, described in (Sabah et al., 1998), helps unpredictable situations
modeling, whereas the Speech Act Theory (i) assigns multiple illocutionary values to the same speech
act thus maintaining ambiguity ; (ii) is more efficient a posteriori than a priori ; (iii) relies on verbs
interpretation by human-based pragmatics, and therefore is difficult to transform into a reliable com-
putational model. The FR theory is closer to an adaptive computational model since it tries to compute
an exchange as an adjustment between locutors mental states. We have adapted this method, originally
designed for human-machine dialog, to artificial agents.
Reasoning. Reasoning, from a learning point of view, is a knowledge derivation mode, included in agent
functionalities, or offered by the 'teacher' agent. Reasoning modifies the recipient agent state, through
a set of reasoning steps. In such a frame, learning is the process that takes as an input the result of a
reasoning procedure over new facts or predicates, and ends up in engulfing them in the agent knowledge
base. Thus, inspired from human behavior, the described model includes the three types of reasoning
described in the preceding section: Deduction, induction and abduction. Currently, our system is based
on axioms which are abstractions of inductive and deductive mechanisms. Abduction is investigated
in some knowledge discovery strategies, and in repair after misunderstanding strategies, that are an
original contribution of this chapter. Abduction implicitly plays a very crucial role. We consider dialog
as an abductive bootstrap technique which, by presenting new knowledge, enables knowledge addition
or retraction and therefore leads to knowledge revision (Josephson & Josephson, 1994, Pagnucco, 1996).
So, its explicit mechanisms, related to 'explanation' in the dialog features, are definitely contributing
to the system task.
Simplification due To Artificial Agents
Transposing a human learner model to an artificial cognitive agent must not be done without acknowl-
edging the particularities of artificial agents. So, although our system is heavily inspired from dialog
between humans and from human-machine dialog systems, it differs from them with respect to the
following items: