SUO: RE: An article on the pitfalls of metadata
Dear John,
I think my take on what you are saying here is the question of
invention vs discovery. I have certainly always thought we were
engaged in an exercise of discovery.
Matthew West
Principal Consultant
Shell Information Technology International Limited
Shell Centre, London SE1 7NA, United Kingdom
Tel: +44 20 7934 4490 Other Tel: +44 7796 336538
Email: matthew.west@shell.com
Internet: http://www.shell.com
> -----Original Message-----
> From: John F. Sowa [mailto:sowa@bestweb.net]
> Sent: 24 August 2003 17:02
> To: West, Matthew R SITI-ITPSIE
> Cc: Richard Cooper; SUO; cg@cs.uah.edu
> Subject: Re: An article on the pitfalls of metadata
>
>
> Dear Natthew,
>
> I'm about to leave to catch a plane, so I can
> only comment very briefly to one point:
>
> > MW: The standards work has been focussed, and low level
> > in that sense, but it has been interesting to note that
> > we have been "forced" by a pragmatic approach to develop
> > an upper level ontology.
>
> I wouldn't be participating in this exercise if I
> didn't believe that a good SUO was (a) important and
> (b) achievable.
>
> But as I said in earlier notes, my major complaint
> about the three top contenders -- Cyc, SUMO, and
> Dolce -- is that they were developed from the top
> down by some human being(s) who drew trees or
> lattices by hand.
>
> I have been arguing since the SUO exercise began
> that the top levels must start with
>
> 1. distinctions, which may be selected by humans.
>
> 2. axioms, which may also be selected by humans.
>
> But the overall structure of concept types and
> subtypes must be computed by systematic algorithms
> (of which there are many -- Formal Concept Analysis,
> FCA, is one well-known example). And the computation
> must take into account the low-level details of the
> kinds of entities that are being classified.
>
> This point has been completely ignored by the
> developers of all three of the candidates.
>
> I have minor qualifications to make on your
> other points, but all of them are related to
> this one big issue: we must derive the hierarchy
> from the bottom up, not impose it by someone's
> best guess. Furthermore, I would emphasize
> that no one's guess -- not even mine -- is
> good enough to accept for a standard.
>
> And as I said many times before, I think that
> your practical experience is important. But
> since your categories were very close to the
> applications, you had solid guidance for your
> selections, even if they were made by hand.
>
> When we are dealing with very high-level
> abstractions, nobody's intuitions are good
> enough to provide guidance. We need algorithms
> for deriving the hierarchy automatically.
> There are several that have been widely used,
> and we should seriously look at them to see
> what they can do.
>
> John
>