Using Misunderstanding and Discussion in Dialog as a Knowledge Acquisition or Enhancement Procecss - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 85

Table 1. Misunderstanding repair common scheme

Step Dialog KB management

Trigger S: say - dissatisfaction P unknown

Launching repair T: askfor - information () Checking known KB implications

…

Asserting knowledge links T: give - explanation ( F ( p ,...)) Increasing KB with F ( P ,...)

Evaluating repair S: say - satisfaction

Table 2. KB conflict repair common scheme

Step Dialog KB management

Trigger & Launching repair S: askfor - information () Checking involved KB implications

…

Asserting wrongness T: give - information ( F alse ) Implication removal

Evaluating repair S: say - satisfaction

In Table 2, a askfor - information () message informs the 'teacher' of the 'student's' KB conflict, and
triggers a dialog headed, this time , by the 'student'. The latter will use a conflict management strategy
to solve the issue. Each 'student' askfor - information () message comes from an introspection on its KB
according to the strategy. The 'teacher' task here is just to give the validity value of the provided informa-
tion. Once the 'student' learns which implication is wrong, it updates its KB by removing this knowledge
and thus solving the conflict. It finally ends this discussion with a say - satisfaction message.
In both situations, each 'teacher's' message conduces the 'student' to analyze and possibly to revise its
KB, the result of this process being then used to answer its interlocutor. This symmetry between dialog
and KB management continues until the issue solving. This end of the discussion allows the agents to
go back to the previous discussion (usually the lesson) and to restore its symmetry.

syste M architecture and iMPle Ment ation

The theoretical approach of the previous section has been specified and partially implemented.

Architecture

The Figure 7 displays the main architecture elements of our tutored learning system. It is composed of
five main structures: The 'teacher', the 'student', the FR, the strategies and the 'World'. 'Teacher' and
'Student' are agents.
The FR are a shared knowledge base about dialogic clauses, to which both agents have access.
The strategies are meta-rules of behavior that help both 'teacher' and 'student' to achieve satisfaction
(positive meta-evaluation) and thus to end the dialog with success (an ending with failure is possible,
since a repeated negative meta-evaluation might appear. Then, the 'teacher' ends up the dialog, because