Information Technology Reference
In-Depth Information
natural order here. It might be possible to use this knowledge, as part of the tree
structure, without requiring more sophisticated natural language or ontology
understandings. For example, every event that takes place, takes place on planet
earth. If we were creating a structured ontology, planet earth would be at the
bottom. Then, for example, a car always drives on a road and so a road should link
to a car, not the other way around. It might be the case that the car branch, when it
gains relations to lots of other things, would be broken off to form a new base, but it
still makes more sense to link from road to car and not car to road. So this ordering,
based on some knowledge of relative size or use in the real world, might also
become part of a structuring rule. It would be useful because the related context-
speci
c information should not be very sophisticated and so it might be possible to
apply the knowledge automatically again. We just need to know that there is a car
and a road, for example. One could imagine a large database that stores different
bands of entities, grouped simply by size or weight, that are not allowed to be
ordered before/after another entity. It is not a typeOf or subClass relation, but a
more functional one. Maybe something like relative use, but it is really only the
ordering that is required. This
fixed ordering would again be a secondary aid, where
the statistical counts and dynamic relations of the parsed text would still have the
most in
uence. The trees of Figs.
1
,
2
and
3
might have their ordering changed
slightly, for example, but the word groups and concept associations would still be
determined by the dynamic text, not fixed knowledge. For example, the mat should
probably be placed before the cat, when the cat branch could be broken off later. It
might be
fl
'
mat
cat
black + sat
'
, or something.
−
−
7 Relation to Earlier Work
This section is slightly different, looking at a speci
c cognitive model, rather than
general theories. It is helpful for developing that cognitive model further and will
hopefully add ideas for a more intelligent system, but can be skipped if the database
model is speci
cally of interest. Earlier research by the author has looked at how a
whole cognitive model might be developed from very simple mechanisms, such as
stigmergic or dynamic links (Greer
2008
,
2013b
). The earlier work described how a
reinforcement mechanism can be used to determine the reliability of linked source
references in a linking structure. These links are created through user feedback only
and are therefore very
flexible, as the feedback can be much more variable than
static rules can accommodate. User feedback adds the intelligence of the user,
which a rule set might not contain. While concept trees are also built from user
feedback, they are then constrained by pre-determined rules and knowledge. They
are also more semantic, complementing the event instances of the earlier work.
A concept tree could therefore be created from similar source types
fl
—
sensor-based,
dynamic input, speci
c concepts, but deal more with the existing structure than the
events that created it. It is still possible to make comparisons with earlier work on a
neural network model (Greer
2011
) that clustered without considering semantics,
Search WWH ::
Custom Search