[topicmapmail] Making ontologies : RDF vs TM

[Murray Altheim]

Vincent Godefroy wrote:
> Hi everybody,
> 
> Is there any documents, white papers on making
> ontologies with xtm ? Google replies me with a
> lot of stuff with RDF/RDFS and OWL.
> 
> How can we see TM in the web semantic effort ?

Vincent,

What developing an ontology basically entails is cataloging a set
of desired concepts and relations, then using and/or implementing
both a document type (for interchange) and tools that can process
such documents. This is basically the same pattern we've seen
since the early days of Artificial Intelligence, and little has
changed in this regard. The "Semantic Web" is just a W3C marketing
gloss over the existing efforts in Description Logics. It doesn't
really touch much upon the wider AI/Knowledge Representation field,
which has a much greater variety of ideas to offer, an entire
multi-faceted community of research, and decades longer experience.

Any graph-based language can be used as an interchange format
for a computer-based ontology. OWL provides a very specific
set of Description Logics semantics. If that's what you need,
you can probably use OWL. If you need more than DL (which most
applications will), you're on your own, i.e., right back to
looking at graph-based languages like RDF, GXL and XTM as a
"carrier" for your semantics. Topic Maps in general provide
higher level semantics than RDF or GXL, which include subject
identity and mapping. These may be very useful to you, depending
on your application.

If you decide to develop an XML Topic Maps based solution, you'd
basically establish a set of Published Subject Identifiers (URLs)
for each concept and relation, then develop tools that can create,
edit, display, manipulate, import and export, etc. documents
containing those PSIs. As above, this is not different than has
been done for decades. Early in 2000 I began by publishing a set
of PSIs for the concepts in the Cyc Upper Ontology. A conversion
of the entire Cyc set of concepts and functional relations would
be possible but very difficult. It'd probably be better to consider
developing cooperative tools, such as hooking Kal Ahmed's TM4J
Topic Map engine up to Cyc or other existing tool sets to provide
TM functionality. E.g., Cyc has no mapping or self-classification
features built in, and would benefit from this.

The place where TM will shine over the alternatives are in the
two areas I mentioned, subject identity and mapping. Identity is
one of the toughest nuts to crack, both technically and
epistemologically. A lot of systems (like OWL) hardly even deal
with identity, and when they do it's often broken. Current
discussions within the TM community have been looking at issues
surrounding identity, and how Topic Maps may provide a clearer
picture and better means of establishing it.

My own research has involved use of XTM as the basis for a set
of ontologies used for authoring. [I have not published any papers
or released any software resulting from this work as of this date.]

You've of course touched upon a subject that no single email can
answer, but hopefully this is a start.

Murray

......................................................................
Murray Altheim                    http://kmi.open.ac.uk/people/murray/
Knowledge Media Institute
The Open University, Milton Keynes, Bucks, MK7 6AA, UK               .

     "Kerry has a positive vision for America, plus the proven
      intelligence, good sense, and guts to make it happen.
      That’s why The Iconoclast urges Texans not to rate the
      candidate by his hometown or even his political party,
      but instead by where he intends to take the country.
      The Iconoclast wholeheartedly endorses John Kerry."

   The Lone Star Iconoclast, Crawford, Texas
   http://news.iconoclast-texas.com/web/Columns/Editorial/editorial39.htm