seaman at noao.edu
Fri Feb 23 09:26:57 PST 2007
On Feb 23, 2007, at 9:36 AM, Ed Shaya wrote:
> Tony Linde wrote:
>>> ordinarily one would include them. For instance, what if it is
>>> like a cataclysmicVariable but not quite a LateType star.
>> Is this not where relations come into play: you could define
>> relations such
>> as 'like a' and 'not quite a' and then, depending on what you want
>> to do,
>> manipulate the objects and relations in your code.
> Yes, that would work. You could allow AstroObject to have a property
> "likeA" or "nearly" which would allow one to toss objects into an
> anonymous class of (F-star, nearly, CataclysmicVariable). I guess
> this is better because you don't create as many named classes, but
> it has basically the same functionally as a
> NearlyCataclysmicVariable. If later on we wish to include F-stars
> as CVs, it would be easy to refactor the ontology.
This usage is similar to VOEvent:
<Why importance="1.0" expires="1574-05-11T12:00:00">
<Name>Tycho's Stella Nova</Name>
<Inference probability="1.0" relation="associated">
A particular assertion of a descriptor (name or concept) can be
associated with several modifiers. The importance and expiration
might be considered as local to the transient alert application, but
the probability and relation are more generally applicable. The
vocabulary chosen for the descriptors can either come from a
controlled namespace like UCDs, or can be expressed in natural language.
This usage was intentionally kept as simple as possible with the
thought that ontologies or other fruits of the semantic labor could
later be harvested when ripe.
>>> whether or not to use strictly observational qualities or
>>> interpreted qualities. As in, what if it erupts and has
>>> emissions like a
>> And this is where the uniqueness of astronomy comes into play and
>> we start
>> to build intelligent *astronomy* applications rather than generic
>> information handling apps.
> Agreed. Although I don't think that astronomy is somehow unique.
> Other fields could benefit from the same thing. But, I agree with
> the gist here that one can create very interesting applications,
> like one that automatically classifies objects by observational
> quantities and then attaches physical attributes/interpretations
> into what the objects are.
Building taxonomies is at the heart of the scientific enterprise.
The late Gary Grasdalen was one of my mentors. I remember a long
ride down the front range in Colorado when Gary presented a
particularly detailed such argument. It happened to center on galaxy
classification, but had wide implications. Two, in particular:
1) Taxonomies (ontologies) are always evolving. The only possible
static taxonomy is one representing a dead science. (I heartily
recommend Richard Dawkins' "Ancestors Tale" for a look inside our own
very dynamic taxonomy.)
2) Each scientist builds and rebuilds her own taxonomies as needed
through the synthesis of all she has previously learned.
Yes, we want and need a shared semantic framework, but it must be
loose fitting enough to permit the very science whose essence it
tries to capture.
Transients occupy one region of phase space where these issues are
especially evident, precisely because researchers of the time domain
wrestle daily with a changing understanding of new phenomena.
Consider the nomenclature that emerges, e.g., "Quasi-Stellar
Object". As newly discovered archetypes are studied, one can
consider that either their classification is constantly changing, or
rather, that an identifier like "QSO" is itself allowed (nay,
encouraged) to move from one place to another in the branches of the
taxonomy. One can find examples of both in the history of astronomy.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the semantics