xml-dev - RE: Soft Landing

["Bullard, Claude L (Len)" <clbullar@ingr.com>]

I agree.  The issue is in the overlaps, what Joshua pointed 
out as the "in the sense of" relationships.  At the point where a 
taxonomy needs a magic relationship to include a term (in that case, 
a description probably to further resric the domain), it is likely 
that it now goes extra-domain and into another.  These ecotones are 
important to identify.  It is likely that the rate of change 
here is high and that can be predictive (intuition).  

Domain experts are essential.  Again, there are concerns about  
authority, credentials, testing, and so forth.  The problem of 
automagic is that techniques like frequency of reference can 
be easily poisoned by superstition and adulterate the domains. 
Human scrubbing is required, the domain should be stable, and 
the access rules determined.  There should also be a reasonably 
good payoff because the definitions have to be maintained.

Now, past the economies, how good would you or anyone else 
say that RDF is for this work?   Are techniques or practices 
emerging for building RDFs similar to those described in 
the Principia Cybernetica or perhaps as Roger Costello has 
been assembling?  There is undoubted power to having a single 
language like RDF for doing this work and that power emerges 
as the domains become stable and interrelated.  I should expect 
the portal builders to be able to take enormous advantage of 
such a polyglot for resource discovery.  Beyond that, in the 
area TimBL describes for business, the routing of application 
messages to such services seems useful on the surface, but I 
have a wait and see attitude.  Coherence, transparency, stability 
and reliability have to be assessed and I don't know how that 
is done yet.  

Len Bullard
Intergraph Public Safety
clbullar@ingr.com
http://www.mp3.com/LenBullard

Ekam sat.h, Vipraah bahudhaa vadanti.
Daamyata. Datta. Dayadhvam.h


-----Original Message-----
From: KenNorth [mailto:KenNorth@email.msn.com]

If I were king of the world, with unlimited budget and unlimited
cooperation, I'd start with a taxonomy and domain experts. Let them define a
domain vocabulary (again I keep pointing to MeSH for medical literature).

Then, when new literature is published each month, run it through machine
analysis to identify new terms that start popping up in the literature
(e.g., XML a few years ago). Also identify relationships to existing
concepts or terms (similarity searches), and so on.  The domain experts
identify an alert level (e.g., 5 citations) and when a term or concept
exceeds that level, it's included in a monthly update they receive -- new
terms and concepts in the literature. They use that information when
updating domain vocabularies on a quarterly basis.

Using a pre-defined domain vocabulary is probably more efficient than doing
it all automagically using inference engines, machine analysis of schemas,
RDF, parsing and so on.

Look at the portals that migrated to a classification scheme, instead of
being simply keyword container searches.