[
Lists Home |
Date Index |
Thread Index
]
--- "Roger L. Costello" <costello@mitre.org> wrote:
>
> I believe that Didier and Mike Champion made mention
> of Google as tool
> which provides semantics in a holistic fashion.
You've definitely captured what I have been trying to
say in this thread better than I have said it. Sure,
there's plenty of scope for "18th Century"
reductionist approaches to semantics, especially in
areas such as medical terminology that have been
rigorously studied since, uhh, the 18th century. I'm
not at all sure that the Web is susceptible to this
treatment (e.g., someone tell me where the xml-dev
mailing list might fit in a taxonomy of websites ... I
feel great sympathy for the people who come here
looking for straightforward answers to their XML
development questions!).
> The critical problem is how
> to create a tool
> which provides semantics in a holistic fashion *for
> computers*. Would
> someone care to take a stab at characterizing the
> nature of such a tool?
Well, I basically dropped out of a Political Science
Ph.D program 20-some years ago because I couldn't
begin to figure out how to address real problems that
traditionally have been addressed in a holistic manner
with the analytic / computational approaches favored
by the social science cognoscenti. I figured that
maybe in 20 years or so the AI people would have made
lots of progress on the technology side, and I would
have made my fortune in the computer biz, and then I'd
return to the subject. Alas, neither scenario has
played out like I hoped, so here I am :-) All social
scientists meet the same fate, someday; cynically
boring folks in an internet cafe. [1]
But seriously .... Nicolas Toper's response mentioned
Zen and the Art of Motorcycle Maintenance, a book I
greatly enjoy [2]. The author spends a lot of time
ruminating on the synthetic and analytic approaches to
understanding, in this case, diagnosing mechanical
problems with motorcycles. I recall him talking about
the analytic/reductionist "scientific method" as
sortof a heavyweight tool that one brings in as a last
resort when a synthetic/holistic approach fails to
solve the problem.
I think that Google may be playing with this kind of
thing (considering that they are buying companies with
ontology technologies) -- use the holistic approach to
sweep up a manageable set of possible matches to a
query, and then crank up the reasoning agents to
winnow them down to the ones that really address the
question. In my other favorite example, a spam filter
might use Bayesian statistics to throw out the most
obvious spams, and then an inference engine might use
semantic/ontological tools to handle the ambiguous
cases in the "MaybeSpam" folder. Another approach
(that Google may also be playing with) might be to
look for higher-dimensional clustering of links across
websites and use them in the ranking of sites. The
obvious example here are the webloggers; there may be
topics on which [insert your favorite popular blogger
here] really *is* an expert and his/her utterances
should be taken seriously, and topics on which he/she
is just blathering and the "me too!" links should be
discounted. It might take a bit of logic to make
these categorizations!
So, maybe one can make rigorous reductionist
approaches more feasible by using wholistic means to
do the "obvious" winnowing down, and conversely
improve the holistic side with a bit of
"reductionistic" logic.
[1] obscure reference to Joni Mitchell's "The Last
Time
I Saw Richard", with the line "All romantics meet the
same fate, someday; cynical and drunk and boring
someone in some dark cafe."
[2] The protagonist/author goes insane in grad school
wrestling with the question of defining "quality", a
question which I believe came up on this list
recently!
|
