|
|
Re: [xml-dev] Mailmen, POST, Intent, and Duck Typing
|
[
Lists Home |
Date Index |
Thread Index
]
Len,
this is very interesting. First time I have come across Grice outside
of academic, linguistic circles. What I had read of his I always
thought it must be applicable to ontology reasoning, but never took the
thought further. It is interesting that the Grician contribution is
classified as pragmatics, the classification Peirce gave his own logic.
Thanks (all) for this thread.
The fact that "dumb" Bayesian
networks with no semantic formalisms have been much more successful
than expert systems in classifying spam, and therefore much more
useful to real people, is perhaps a beacon in this regard.
There are those who attempt to combine the two (losing the "purity" of
both), each node of an ontology tree computed against a statistical
algorithm.
But the intriguing thing about statistical analysis is that in some way
it is not "dumb", it really is an open question as to how neural type
networks map into brain/human social functioning. Stochastic process
and models of these processes are often givens in psychological
research, i.e. a neural net model may be taken as sufficient to model
peripheral processes to that under investigation.
Ontologies are convenient ways of organising information that take some
of their convenience from the fact that their structure contains
information. But there is no reason to believe that because an ontology
can be generated it is a discovery of what already exists, on the
contrary, it is an intellectual invention that provides short cuts to
implied knowledge in some circumstances. C.S. Peirce demonstrated the
logical necessity of the underlying relationships, not particular,
specific ontologies.
I think that the issues are not of the complexity of the machine, but
the complexity of the user if the user is human. Methods that may work
for machine <-> machine negotiation may not work for human
<-> machine, pragmatically speaking. I think this is an area for
research and clarification.
Adam
On 24/02/06, Bullard, Claude L (Len) <len.bullard@intergraph.com> wrote:
If I make a bet on the cat being dead, does that
alter the probability, the fact, or in any way
change the need to open the box and look?
On the other hand, if I am making a bet on
spam, my risks are lower than the cat betting
that I am going to open the box.
Given the frequency of spam, the occasional
misclassification is a low cost event, strictly
speaking although there is a probability that
I will miss something important.
Pragmatic systems are learning systems.
len
From: Chris Burdess [mailto:d09@hush.ai]
The fact that "dumb" Bayesian
networks with no semantic formalisms have been much more successful
than expert systems in classifying spam, and therefore much more
useful to real people, is perhaps a beacon in this regard.
|
|
|
|
|
|