Information Technology Reference
In-Depth Information
to check if the new knowledge doesn't generate a conflict in its KB. If this happens, it will have to use
a conflict management strategy.
dialogue Strategy
Each learning situation requires a strategy to guide the dialog. Moreover, interaction may lead up to
misunderstanding and conflicts during the dialog, each one requiring a specific strategy. We begin by
describing the basic strategy for the lesson, then we deal with the 'curious student' strategies, and finally
with the problem resolution strategies.
Lesson Strategy.
Our strategy is based on the socratic dialog principle: The 'teacher' will provide each
piece of knowledge (formulas in our case) to the 'student' and will wait for its satisfaction or dissatisfac-
tion before continuing the lesson. By this way,
each teacher's message is at the same time a knowledge
gift and a question about its understanding, thanks to the FR
. As long as the 'student' is satisfied, the
'teacher' continues providing each formula until the end of the lesson. If the 'student' expresses its
misunderstanding, then the 'teacher' will have to use a misunderstanding strategy, which we describe
later in this section.
KB Connexity Increasing Strategy.
At any time, the 'student' can ask itself whether there could be
a direct link between two pieces of data of its KB which are currently not linked (belonging to two
distinct KB components) or indirectly (by a long path).
Let's take the example of predicates it learns: For instance, the 'teacher' teaches the two next for-
mulas:
human
(
x
)→
mortal
(
x
) and
human
(
x
)→
animal
(
x
), the 'student' will be able to ask itself about a
possible direct link between
mortal
(
x
) and
animal
(
x
). The dialog will be then used to ask to the 'teacher'
if it own such a relationship.
The 'student' can then proceed to the two following questions, each one addressing a possible link
between predicates:
•
askfor
-
information
(
mortal
(
x
)→
animal
(
x
))
•
askfor
-
information
(
animal
(
x
)→
mortal
(
x
))
The 'teacher' could, for example, confirm the second relationship to the 'student'. The student's KB
has then been enriched with a new relationship.
Predicate Base Widening Strategy.
Each time the 'student' integrates a piece of data it can use it to
generate new knowledge. Thus, if it learns a formula of type
p
(
a
) where
p
is new but
a
already known
in another formula of type
q
(
a
), then it can ask itself whether there is a link between predicates
p
and
q
. It can then question the 'teacher' about the validity of
q
(
x
)→
p
(
x
).
It's a kind of induction since, from facts having one thing in common (the constant), the 'student'
wonders if there is a rule linking them. If the formula is true, then the 'student' will be able to deduce
that every constant which satisfies
q
will now also satisfy
p
, i.e. increasing the base of the predicate
p
.
The dialog of the Figure 2 illustrates the use of this strategy. The KB evolution can be followed, step
by step, in the Figure 3. We can note that the curiosity of the 'student' has also enabled it to preserve
its KB connexity by adding the implication (Step 3).
Search WWH ::
Custom Search