Information Technology Reference
In-Depth Information
This, naturally is the anchor of a revision based process, where the a given concept, playing the role
of an assumption, is subject to confrontation with the inner knowledge source of the requiring agent.
Thus, it drives the latter to proceed to derivation (by reasoning).
However, here appears the question of evaluation: Is the provided knowledge reliable? In a general
situation, one will tend to implicitly accept Grice's recommendations about information quality: Agents
must not utter something unless they believe it to be true. But are their beliefs equivalent to a certain
knowledge? Is any interlocutor a reliable knowledge source?
Some authors try to meet this issue by assuming that a consensus about knowledge is what could
be the closer to truth. It is what happens in negotiation-based models for knowledge revision. Mutual
belief revision, defended by (Zhang et al., 2004), reaches an equilibrium assumed to be a clue for the
quality of shared knowledge. But negotiating knowledge among several agents might tend to favor group
attitudes instead of a search for truth. Several agents sharing the same wrong belief might weigh more
than one proposing a good but individual solution. Therefore, these questions have led some authors
focus on dialog between two agents, one playing the role of a reliable source (and being acknowledged
as one), and the second, playing the role of the learning agent (Beun & Eijk, 2003). However, these
dialog situation are assimilated to game strategies in the cited paper. It is also the case in (Amgoud &
Prade, 2003). Therefore, sometimes dialog is more an interaction process than a real 'logos' (the Greek
root in the word dialog, i.e. related to discourse and language) process.
dialog in Learning
Human learning with human teachers relies on language and communication. A classroom learning in
which a teacher talks, and students sometimes ask questions is not a dialog situation, and obviously it is
a personal teacher, with a true dialog between teacher and student, that gets the best results in knowledge
transfer. The situation could be rapidly described as: A teacher provides knowledge when asked, or as
a lecture. The feedback commonly observed in natural dialog is that the teacher, playing the role of the
knowledge source could be addressed by the learner, in order to test whether the acquisition process
has succeeded. This is called
tutored learning
.
Human learning with machines has tried to emulate this privileged interaction. Authors in the domain
have repeatedly insisted on the importance of dialog as an acquisition and evaluation process for the hu-
man learner (VanLehn, Siler, & C.Murray, 1998), (Muansuwan, Sirinaovakul, & Thepruangchai, 2004).
Literature is truly abundant on the subject and we cannot but grossly summarize the main tendencies
of the state-of-the-art. The main achievements could be described as:
•
Dialog models in computer science are based on
intentions
(Cohen & Levesque, 1992), and define
dialog instances as a succession of planned communicative actions modifying implicated agents'
mental state.
•
Thus a dialog-based ITS has a dialog model implemented in it. The ITS is in a knowledge transfer
situation, and as a goal, needs to teach a set of knowledge items. On the other hand, the human
partner also has a goal: He/she needs to learn these very items. The ITS must check that the human
has effectively acquired knowledge.
•
Several researches have shown that a predetermined plan does not work, because the human's
actions and/or answer cannot be predicted. Therefore, ITSs react
step by step
, according to the
Search WWH ::
Custom Search