Information Technology Reference
In-Depth Information
conclusion
This work was an extension of previous studies in which the rationale is the following: Sole communica-
tion could be a knowledge acquisition process as efficient (if not more) than other gathering mechanisms,
for artificial cognitive agents. This communication needs here to be directed and controlled by goals, and
it can be instantiated into the sophisticated shape of a dialog. Therefore, we have centered our research
on studying equivalency relationships between dialogic structures and knowledge acquisition and revi-
sion process. Doing this have led us to restrict the study to the typical case of a socratic lesson, in which
two agents, one playing the role of a teacher and the other the role of a student, interact by exchanging
messages related to a set of knowledge to be taught. This case has been qualified as a representative
tutored learning method. It advantages are:
•
Learning is a major incentive for one of the agents, and teaching is symmetrically the teacher's
one. In this case, student agents are naturally willing to undertake the acquisition process, and
teachers are also willing to provide true and reliable information (highly cooperative situation).
•
Since the teacher must deliver true information, this restricts its offer to its skill domain, that
is, information for which it is able to design a lesson. At the same time, this makes it a reliable
knowledge source for the student, which will accept any statement from the teacher as true.
•
Knowledge transfer (and acquisition) is the main goal of both agents, and uses 'discourse' only. By
avoiding mixing actions and messages, we isolate the dialogic aspect and therefore may examine
its impact much better than in any other situation.
•
Lessons are very interesting situations in which several cases of knowledge management appear:
Enhancement of facts (predicates instantiated with constants), of concepts (predicates) of laws
(implications); Creation of new implications (inductive process); Implications retraction; Facts
associations to predicates which do not contain them in their validity domain (abduction).
•
At the same time, lessons allow conversational structures to reflect and handle all these cases
may occur. This allows us to follow through dialog the track of knowledge management and thus
refines our assumption about a possible symmetry (if not exactly an equivalence) between dialog
and reasoning.
The study of such a constrained situation has led us to define a notion of
connexity
for a knowledge
base (KB), allowing to assess the connection level between each element of knowledge of an agent and
so to give it two new goals: Increasing its KB connexity, as well as optimizing its content. As the dialog
situation in highly unpredictable and may follow no previous plan, we have adopted the functional role
(FR) theory to easily model dialogical exchanges. Agents use strategies to learn new knowledge and
solve conflicts between external and internal data. (Angluin, 1987) tackles the problem of identifying
an unknown subset of hypothesis among a set by queries to an oracle. Our work differs mainly in:
•
The communication mean:
We use imbricated dialogs instead of queries
•
The learning's aim:
Our agents aim at learning new formulas and increasing their KB connexity
instead of identifying assumptions
Moreover, it seems that the FR theory has possible patterns for typical communication occurrences:
Misunderstanding and Discussion. As developed in the paper, misunderstanding, limited to the student
Search WWH ::
Custom Search