[topicmapmail] Making ontologies : RDF vs TM
Murray Altheim
m.altheim@open.ac.uk
Sat, 23 Oct 2004 19:46:40 +0100
Lars Marius Garshol wrote:
> * Murray Altheim
> |
> | It was rather quickly discovered that these fixed categories didn't
> | work very well for the real world, that almost nothing is *about* a
> | single subject, that the whole idea of "about" is very
> | context-based. So when in computing, say as in RDF, we see
> | "rdf:about", or in Topic Maps we see "subjectIndicatorRef", we
> | should be very suspicious, or at least very careful. Nothing has or
> | is just one subject. The statement of subject-hood is contextual.
>
> I think you are right that the notion of single, absolute, discrete,
> and objective subjects is very dubious, and in fact it was Bernard who
> persuaded me that this is the case way back in Orlando in 2001. He has
> been talking about this for a long time, and I think he is right.
Yes, I think Bernard and I fundamentally agree on this issue. But he
and I probably have more talking to do about this.
> However, the question isn't really whether the notion is valid or not,
> or whether it applies in the way we think it applies, but whether we
> can make it work. If my IT system works the way I want it to, then who
> cares, really, if it fails to be an exact model of the "real" world?
Well, of course. That was precisely my point about this problem not
having any real impact so long as people had a correct understanding
about what was actually *happening* in the Topic Map, i.e., not a
declaration of universals but a highly contextualized form of
communication, valid only within its context. If it works the you
want to that is by definition a form of validation, but it might not
work for anyone else in the "real" (uncontextualized) world. Quoting
Patrick Brézillon, who I mentioned in my previous message, begins [1]
with:
Brezillon defined context as a collection of relevant conditions
and surrounding influences that make a situation unique and
comprehensible.
[Jack, this is the very quote I'd point you to as a summary of what
I'd been getting on about in our own discussion, i.e., its explanation
in the rest of Brézillon's paper and others.]
In contrast to Sowa's Conceptual Graphs, Brézillon describes the idea
of a "contextual graph" [1][2]. If Topic Maps are not already exactly
that, I think they're likely the ideal substrate upon which to construct
a contextual graph. Admittedly, Conceptual Graphs have an ability to
wrap statements in layers of context too, and I'm not yet clear on
exactly how they differ (if they do) in their underlying models. But
from the perspective of the language used to describe each, and the
overall emphasis on context provided by Brézillon, I've been spending
more time lately on his notion of contextual graphs. In particular,
trying to get my head around how his notion of "explanation" [3][4]
is backed up mathematically, and how this might be different from
alternatively-named similar ideas in both logic and epistemology. (If
only I had enough time to cover this one subject alone in my reading!)
> Ann had an interesting insight about this at the reference model
> workshop in Montréal this August, which was that if one didn't do any
> merging of topics up front, one could apply different theories of
> subject identity to them at will, which are effectively equivalence
> relations that form one equivalence class for each (hypothesized)
> subject. (This is my understanding of it; hers may differ. :)
>
> I thought this a very interesting idea, and one that I think in part
> overcomes this problem. I'm not sure where it will go and what will
> happen to it, but for me this clarified a number of things.
I can't speak to the context of Ann's statement, but from what you've
provided it *seems* to be making some kind of up-front assumption of
the purity (author's intention?) of the Topic Map prior to any sort
of local merging. It doesn't *seem* to solve this problem at all. How
do you see it as, in part, being a solution? I don't myself see it.
> | (I use Topic Map termino- logy here because I am also emphasizing
> | that I think we can still use Topic Map technology to do information
> | or "knowledge" modeling; we just need to alter our approach a bit
> | when talking about subject identity.)
>
> How do you think we need to alter our approach? (I'm asking because I
> couldn't find an explanation of this in your email. Which is not to
> say it may not have been there. :-)
Simply by being very clear about context, i.e., that an individual
Topic Map is not a statement of universals. By way of example, if
one looks at the OWL specification (or many, many others), it's
not stated anywhere that OWL is a very specific set of semantics
*within a specific community, for a specific purpose*. There is,
at least to the unenlighted, the idea that OWL's subclass, for
example, is a subclass suitable for all uses of "subclass", as if
one didn't even need to delve into how Description Logics, or the
W3C-specific version of DL-within-OWL defines class, and that whole
recursive problem of needing to further define the words used in
the definition of class, etc. etc. etc.
In other words, OWL acts as if it has some sort of universal
semantic validity. And 90% of those using OWL won't run into any
problems because they would be, like you said earlier in this
message, their IT systems just seem to "work." Problem is, of
course, that OWL was designed as a universal, so when your IT
system talks to someone else's there's no way of really capturing
the contextualized differences between the built-in, contextualized
implementations, such that errors or problems that will occur (even
if they're not noticed until someone's bank balance is wrong, a
military ordnance is sent to the wrong place, or you receive a
sex toy in the mail instead of a baking tin). Hmm. Might not be a
problem for some people, but point is, the kinds of errors that
crop up might not even be noticed, which is more a problem than
if they are. People tend to filter out incongruities in understanding
between each other when communicating.
> | The problem is that almost the entirety of western culture is built
> | upon thousands of years of thinking about things as subjects, as
> | categories, as identifiable "types", when I believe reality is
> | telling us otherwise.
>
> I agree.
>
> | My feeling is that it isn't crackable, that we'll never have any
> | kind of universal ontology because there are almost no shared
> | universals amongst all people.
>
> I agree again.
Perhaps the only thing we *really* need do is emphasize context and
make sure we include it and use it appropriately. PSI sets are
published not as universal ontologies but as agreed-upon statements
about a domain by an individual or group, within a particular
domain of time, community and purpose. It may be that our notion of
scope is weak in implementation, but OTOH, there's no inherent
limitation in our ability to construct complex relationships
creating context, such that some "collection of relevant conditions
and surrounding influences that make a situation unique and
comprehensible" could be reified as a Topic and used to scope a
specific Topic Map construct. It's not "a cat is on a mat," it's
"a specific cat is on a specific mat at a specific time", with the
devil in the details defining "specific", and we're still leaving
out C.S. Peirce's Thirdness, which is where we begin to derive
meaning, in contextualized human interpretation, not in the statement
itself.
So another way of putting this is to take every instance of a logician
using the "universal quantifier" (i.e., "for all") and add a wrapper
of context around their head, like a big Turkish towel. There is no
and can be no universal quantifier, despite what every FOL advocate
might believe.
Murray
---------------
[1] "Context Dynamic and Explanation in Contextual Graphs", Brézillon, P.
http://www-sysdef.lip6.fr/~brezil/Pages2/Publications/26800094PB.pdf
[2] "Operational Knowledge Representation for Practical Decision Making",
Jean-Charles POMEROL, Patrick BREZILLON
http://www.hicss.hawaii.edu/HICSS_34/PDFs/DTISA02.pdf
[x] "Explanation as Contextual Categorization", Leslie Ganet, Patrick
Brézillon, and Charles Tijus.
http://www-sysdef.lip6.fr/~brezil/Pages2/Publications/26800142LG.pdf
[x] "Modeling and Using Context: Past, present, and Future", Brézillon, P.
http://ftp.lip6.fr/lip6/reports/2002/lip6.2002.010.pdf
......................................................................
Murray Altheim http://kmi.open.ac.uk/people/murray/
Knowledge Media Institute
The Open University, Milton Keynes, Bucks, MK7 6AA, UK .
A schoolteacher at a Bush event was on her way to the bathroom when
she was stopped by a volunteer and told she wasn't welcome. The
volunteer pointed to her T-shirt and said it was 'obscene'. She and
her two friends (also wearing the same shirts) were escorted out by
police officers and threatened with arrest if they did not comply.
The T-Shirts read: 'Protect Our Civil Liberties'"
Shutting Them Up
http://dneiwert.blogspot.com/2004_10_17_dneiwert_archive.html#109830472470609571
Dismantling Democracy
http://dneiwert.blogspot.com/2004_10_17_dneiwert_archive.html#109830622308006215