thomas at astro.umd.edu
Sat Sep 15 08:10:15 PDT 2007
On Friday 14 September 2007 10:55:55 pm Doug Tody wrote:
> Although I normally argue for keeping it simple to start with, I don't
> think I agree with this. Yes, broader and narrower are relationships
> (as is hierarchy), and we already see this in the proposed SV and in
> other things like UCDs. The thing is, when we classify a particular
> object we would like to be as precise as possible. But when we
> search for objects as candidates for analysis we would often like
> to be able to express a broader category which includes all the
> subcategories, and refine it iteratively based upon the results
> of a search. Hence for any useful astronomical vocabulary it is
> necessary to express relationships and be able to map one to many,
> to several levels of detail.
I recall that Ed did an analysis of the various extant astronomical
ontologies (each by a present member of this group!) a while ago
and commented that it was striking that there were many similarities.
Just as striking where the differences, and these would forever, it
seems, prevent eventual mergers (last sentence: my words not Ed's).
My point here is that certainly there are relationships we may all
draw between classes of objects and there will be little argument.
However, as we reach for greater specificity, there will develop
greater and greater resistance to various categorizations, which I predict
will eventually bog the whole process down. This is why I vote for
simple enumeration of the terms of interest to start. The power of
these standards flows, in part, because we largely all agree with them.
Having a simple list to start will be a good way to build/reach consensus,
and have the bonus of this (new post-UCD) group actually producing
something of use to the rest of the VO community.
I agree that once in place, the work of relating the terms should proceed,
and will (probably) show a further list of missed terms. This stage should
proceed carefully. I don't know where the line is to be drawn on relationships
(one can be awefully specific in an ontology) but it needs to be on the side
of less specific, more broad. Otherwise, we will descend into arguments
over minutiae, and it will be hard to deliver a 'product'. I continue to believe
that where really specific, powerfully described relationships are needed,
a special ontology will be developed by the domain experts who need it, and
perhaps there will be developed a mechanism by then whereby they may
share this back with the community.
More information about the semantics