[topicmapmail] Intuition and formalism

[Dean Allemang]

While discussing a paper that I have collaborated on
(http://www.topquadrant.com/documents/TQ03_Semantic_Technology_Briefing.PDF),
I sent the following to Are Gulbandsen, who suggested it might make an
interesting topic for this list.  The original is a personal letter, rather
than a list posting, so please forgive the informal tone (and the length of
the comment).  Here goes:


It seems to me that RDF and Topic Maps have a very strong difference  in
philosophy stemming from different basic assumptions that were made  very
early in their development.  The way I see RDF, is that its  inventors had
the goal of taking a well-known knowledge representation formalism
(object-attribute-value triples), and embedding it in a logically
consistent way into a full descriptive system.  Because of their  background
in logic, these inventors made the development of a  model-theory based
semantics of RDF (http://www.w3.org/TR/rdf-mt/)  high on their priority
list.  Because of this formal foundation, it is  possible for formal
logicians to make very detailed comparisons and  critiques of the semantics
of RDF and, say, OWL (eg.,
http://www.cs.man.ac.uk/~horrocks/Publications/download/2003/p50-horrocks.pdf).
For a formalist, it is impossible to take any language seriously until this
kind of background work has been done.

Topic Maps, on the other hand, as I see it, began with an awareness of  the
sorts of needs that people have when they model things.  Topic  Maps began
with an awareness that there is a difference between the  map and the
terrain, and thereby named two entities, the "Subject" and  the "Topic" to
handle this difference.  The Topic Map inventors also  realized that there
is a difference between either of these things and  the electronic (or
other) artifacts that mention them, so there was a  third thing called an
"Occurence" to handle this.  And so on - Topic  Maps were not developed so
that they would have particular formal  properties as their main objective;
rather, Topic Maps were invented  to respond to certain modeling needs.  As
far as I am aware (if I am  mistaken here, I would be happy if someone would
 point me to this work),  there is no work on model-theoretic formal
semantics of Topic Maps as  there is for RDF.  I don't think that this is a
surprise, given the  goals of Topic Maps.   Nor do I think that this is even
necessarily a  lack on the part of the Topic Map standards and research;
after all, logicians do not make up a very large part of the marketplace,
and the commercial advantages to being able to perform analyses like the one
I reference above are not clear to me.

In the paper I mentioned at the beginning of this, I tersely summarize all
this with two overly brief comments:

o RDF has very formal connections,
o while Topic Maps have intuitive ones

In defence of these comments, I do believe that more comprehensive and
extensive work has been done on the formal semantics of RDF than has  been
done for Topic Maps (again, please correct me if I am mistaken),  and I also
believe that the Topic Map standard is much more intuitive  for most
modelers (except those who really wish that they were writing  logic
programs instead of models).


Dean