Lists Home |
Date Index |
On Wed, 10 Nov 2004 09:15:28 -0500, Michael Champion
> On Tue, 09 Nov 2004 20:29:44 -0800, Tim Bray <email@example.com> wrote:
> > On Nov 9, 2004, at 9:18 AM, DuCharme, Bob (LNG-CHO) wrote:
> > > the amount of practical, usable RDF data still seems
> > > remarkably small.
> > Just a reminder that the RDF.net challenge is still open: see
> > http://www.tbray.org/ongoing/When/200x/2003/05/21/RDFNet
> FWIW, my recent epiphany that the Semantic Web  stuff is not as
> wacky as I once thought came partly through the possibly heretical
> thought that there is no need to convert much actual data, and perhaps
> not much metadata, into RDF. The powerful Semantic Web idea for me is
> the notion of an ontology -- meta-metadata if you will -- that relates
> concepts and relationships in real-world metadata (service
> definitions, XML schemas, data dictionaries, etc.) to one another in a
> way that supports automated inference.
<snip>relationship of RDF to Ontologies discussion</snip>
> Of course, one logically could build the Semantic Web from the bottom
> up by making the actual information out there on the web, in
> databases, etc. RDF-compatible so that generic reasoners could follow
> the chain of resource definitions back to something that an
> application understands. I'm just predicting that it is far more
> likely, and far more practical in real enterprise projects, for
> applications to gradually migrate the logic for finding and relating
> data and metadata from being hard coded to being inferences over an
I agree but I'd also qualify this:
I agree because I believe the way metadata becomes truly useful within
an organization is when it is exposed and manipulated and experimented
with and validated. In other words, when organizations can treat
meta-data just like any other data it gives them a whole new way of
managing the organization. A good way to flexibly manipulate metadata
is with rule based systems. The rules themselves can then be treated
as metadata and you get to recursively refine both the metadata and
the way you manage it. (Comments on the why is recursion difficult
thread in a moment :-).
The qualification is that this is currently expensive in terms of
hardware and people skills. The fact that generalized graph traversal
is NP complete (with all the usual qualifications about what that
means) tells us that generalized manipulation of metadata managed as
RDF not going to become a tractable approach for really broad problem
domains any time soon. (Your footnote applies.)
This thread keeps bumping up against this; the issue of why it is hard
to transition from a local Ontology to a global Ontology has direct
relationships to some of the fundamental issues in Computer Science.
All the hand waving in the world isn't going to make this go away:
unless there are some fundamental discoveries there are certain
problems that can't be solved in reasonable amounts of time with
reasonable amounts of resources (Moore's law only helps expand the
definition of reasonable). The wild card is quantum computing and the
like, but if the success of the global Semantic Web depends on that
coming to fruition, then once more I have to conclude that the general
solution isn't going to be packaged up as a commercial application any
> That has a couple of advantages: the extreme ugliness of RDF syntax
> is not a problem for very many people, and there is a plausible
> evolutionary path from the world of today to the vision for tomorrow.
> After all, any domain that is well defined and stable enough to be
> even plausibly managed with hard coded relationships is obviously well
> defined and stable enough to be modeled with an ontology. Using
> semantic technology improves the ability to accommodate change and
> diversity, using an explicit model in a single language rather than
> multiple implicit models in various programming and database
> languages. Furthermore, one can use conventional RDBMS and XML
> technology (and conventional application code) to manage all the data
> and existing metadata, the only information that potentially would be
> managed with exotic triple stores and new query languages is the
> ontology itself.
> Anyone want to point out flaws in this assessment?
I don't see any real flaws. It's how we're attacking the problem
(with the exception of the Ontology which we don't have) and it works.
We're doing research that saves peoples lives, paying the price for
the hardware and people skills can be justified if you can attack
problems that you couldn't attack before.
I believe local Ontologies and limited knowledge discovery will work.
Organizations (like the government) that can enforce broader standards
will have more success in extended the success of such systems into
new areas. I also believe there will be some limited success with
ad-hoc discovery within very narrow vertical domains. The value here
won't be new knowledge but the discovery of shared interchange
capabilities: Eg. my system just determined that your <postcode> is
equivalent to my <zipcode> (and in doing so it found a way to
translate between the two) so I can now exchange address information
with you without writing any new code. For the humans involved the
information that we can translate between zip code and postal code
isn't news. However, the capability to do this at the machine level
without human intervention has real value.
>  Apropos John Loutzenhiser's comments in the parent thread, I'm
> talking about applying the technologies that come out of the Semantic
> Web activity to relatively constrained and static domains for which a
> top-down formal semantics approach is appropriate. I agree that
> applying them to the Web as a whole will be a "bloated, brittle,
> bitter disappointment".
In other words, I probably didn't really need to make any of these
comments, but when has that ever slowed down the discourse? ;-)