Information Technology Reference
In-Depth Information
deals with the problem of micro-states, where possibly Fig.
4
has only one and
Fig.
5
has two, but Fig.
1
is better because the whole dataset is more coherent and it
is already at its minimal state.
6.3 Language Structure
Another comparison should be with how we process natural language. Language is
so
fluent that there are not many restrictions on what can be said or written. As the
concept tree is a simpli
fl
ed model of natural language however, it might allow some
rules to be included. Generic or autonomous rules are desirable and also plausible.
They might be thought of as an extra layer above the basic statistical counts that
help to direct the initial structure. They would not be allowed to override the
triangular count rule however. The ordering used in WordNet
5
(Fellbaum
1998
;
Miller
1995
), for example, is the sort of ordering that would be useful. The base, for
example, could typically be formed from nouns and verbs, with adjectives or
adverbs forming mostly the leaf nodes or end branches. In a real-world sense, the
descriptive words would possibly de
ne speci
c instances of the more grounded
'
'
noun or verb concept groups. For example,
, gives a
count of 1 initially to each concept and so the ordering before adding to a tree could
be changed. The rule might state to add
the black cat sat on the mat
'
cat
'
at the base instead of
'
black
'
,asitisan
object. Then possibly some sort of reverse polish notation
'
cat
—
mat
—
sat
'
to push
nouns down, or just
. So the exact language structure might get lost,
but the associations will still exist and the rules will help to reconstruct text
sequences from the tree. As another example, we can have a cat and a mat, but
maybe only the
'
cat
—
sat
—
mat
'
mat, and so on. Descriptive nodes at the
end would also help to relate the concept tree more closely with earlier work, as
described in Sect.
7
.
'
black
'
cat sat on a
'
red
'
6.4 Natural Weight
The following, associated with size or weight, is possibly even more interesting. It
would be a strange way of looking at ordering text, but it is again a physical-world
rule being applied in a slightly different context and again relates to the idea of sub-
concepts. Note that text often relates to real world objects and so its construction
would have to be consistent with the physical world. In the real world, it is often the
case that the largest and therefore heaviest entity, resides at the bottom of things.
Putting a heavier object on-top of a lighter one is not often done and so there is a
5
I have to note my recent interest in WordNet, although, most of the new theory here was
formulated before that, with WordNet then supporting it.
Search WWH ::
Custom Search