some semantic puzzles from VOEvent

[Bernard Vatant]

Some suggestions of tracks to answer Roy's puzzles ...

(1) Adding an interpretation to a VOEvent. In the VOEvent schema, we have a "Why" section
where the publisher can say what they think the event is (supernova or an outburst of a
variable star or a microlensing event or GRB etc etc). We would like a formal vocabulary
so that, for example, the computer can understand that "Supernova" and "SN" are the same
thing. However, my simple attempts to do this as a new branch of the UCD tree blew up into
a storm of controversy. How can we make progress here?

BV : Preliminary question : I've mined the whole IVOA website looking for VOEvent schema,
but could not find it. Is it available somewhere? Even as a draft?
Seems to me that a science ontology should always make distinct the representation of the
*experimental data* (which if I get it right, are what VOEvent(s) are about), and
*interpretations* of those data. The word "event" is overloaded here, since it seems to
encapsulate both, hence the difficulty, frequent in science. Although everyone involved in
experimental science is aware (or should be) in theory of this distinction, in practice
it's often forgotten when there is global implicit agreement on interpretation. When I say
: "This is an infrared image of the planet Jupiter" I somehow forget the distinction. I
should say : "This is the result of a process in which X have gathered bytes using such
telescope, CCD, software ... and that Y interprets as being an image in IR of the planet
Jupiter". But nobody would do that in that case, because everybody knows (or thinks so) or
at least implicitly agrees upon what Jupiter *is* and looks like, what a planet *is*, and
so on. There is a unique (implicit) interpretation in the common (implicit) ontology. Of
course, this breaks for SN or GRB, where the question of interpretation is critical. I've
an old Web Page about that issue [1].

If one wants to clarify and formalize, this distinction has to be captured in the
ontology. The way I see it, roughly and non-exhaustive, and independently of specific data
or even scientific domain (this applies also in Biology, Earth Sciences, whatever). An
"Event" is composed of one "Fact" and one or more "Interpretation(s)". When I say "Fact",
read "experimental fact".

Components of Fact
	What : Data Set (including values, units etc ...)
	When - Where : Space-Time location of the experiment
	How : Tools (Method, Instrument, Protocol ...)
	Who : Actor in the Fact (observer, data curator ...)
	Related Facts : Previous, Follow-up
	Framework : Research Program ...
	Why : Purpose of the event (why the data have been gathered for, hypothesis, whatever)
	...

Components of Interpretation
	Science Object : The thing(s) data are about, and their type
	Model (this is more tricky : includes theoretical framework, world model, ...)
	Who : Who makes the interpretation
	Publication : Various material documenting the interpretation
	Support : Other facts and interpretations contributing to the interpretation
	Similar : Other facts interpreted as being about the same object
	...

I would be happy to propose a more formal description of all that in a OWL format if folks
are interested to review it.

(2) Suppose we have a collection of Event observation packets *that are all about single
astrophysical event*, but made by different people from different telescopes. How can I
look at the interpretations ("Why" sections) and decide if there is a difference of
opinion, or a change in interpretation with time? One report says the Event is a
Supernova, another says it is a SupernovaType1a, another says it is a SN and another says
"bright source associated with a galaxy". Are these four in agreement?

BV : I've underlined *that are all about single astrophysical event*. Think about the
meaning of this sentence at the light of the above distinction. You have different
experimental facts, interpreted as being about the same object. Observer X has taken an
image later interpreted as being a SN 1c in NGC 4038. Other, looking to the same sky
coordinates, have obtained other data that they interpret as being images of the "same"
object in the "same" galaxy (whatever this "sameness" actually means ...). The "object" is
named 2004gt. Looking at some description :
http://www.rochesterastronomy.org/supernova.html#2004gt ... the distinction between the
fact and the interpretation is far from obvious.

Now to answer directly your question about agreement, seems to me that there are three
levels of agreement:
1. Agreement about speaking of the same fact(s) - same observation(s)
2. Agreements about interpretation :
	2.a : Agreement on "sameness" of object across different observations
	2.b : Agreement on type of this object
For all those, all the point is to explicitly identify the subject of conversation.
Agreement between SN and SN 1c interpretations can be inferred if the ontology of objects
declares that SN 1C is a subclass of SN.

(3) In VOEvent, there is a section for saying what was measured, and it is a collection of
parameters that are characterized by UCD. For example PEAK_COUNT=3243 and R-MAG=17.5 and
BANANA_FLUX=4.654. If I have a lot of VOEvent packets that refer to the same astrophysical
event, can I extract a SED or light-curve (or time-dependent SED) by federating all the
separate observations?

BV: The answer is "yes" if the "what was measured" refers to an interpretation with
agreement on 2.a (in terms of object sameness). When I extract a light-curve I am in the
framework of interpretation. I mean there is no light-curve that is a fact, unless it's
directly captured by a single instrument, protocol ... As long as you put together data
from different sources, the agreement about those data being about the same thing supposes
an (there again, most of the time implicit) agreement on a world model where this object
is supposed to "exist".

This seems somehow independent of any agreement on 2.b (type of the object) although it's
not obvious that there can be any agreement on the sameness of an object if there is
disagreement on its type. From a formal viewpoint, suppose you have declared in the
ontology that SN 1a and SN 1c are disjoint subclasses of SN, if interpretation X says that
E is instance of SN 1a and interpretation Y says that E is instance of SN 1b, but agree
that E is the same object (same object identification), I can infer that X and Y are
inconsistent (read : disagree).

Hope that is helpful, and not adding to confusion :))

[1] http://perso.wanadoo.fr/universimmedia/subjects.htm

**********************************************************************************

Bernard Vatant
Senior Consultant
Knowledge Engineering
bernard.vatant at mondeca.com

"Making Sense of Content" :  http://www.mondeca.com
"Everything is a Subject" :  http://universimmedia.blogspot.com

**********************************************************************************