Context-Sensitive Semantic Relations

Dear SKOS community,

I would appreciate advice on a question about extending SKOS'
functionality.  I work with the InPhO Project (http://inpho.cogs.indiana.edu),
and we are currently considering a transition to SKOS.  We are particularly
attracted to the fact that the SKOS:broader relation allows one to be
agnostic about transitivities.  However, we would really like to infer
*some* transitivities--but only ones that are context specific.  Let me
explain.

A primary component of our knowledge base is a taxonomy of philosophical
ideas, developed and populated by a combination of statistical measures on
a corpus of digitized text and expert feedback.  Most basically, we use an
informational J-measure to decide when two ideas are related, and an
entropy-based measure to decide when one idea is more general than another.
 This allows us to build and populate a taxonomy of philosophical ideas,
organized by their semantic relevance.

What is really desired is a transitive skos:broader relation, but one that
holds only in certain contexts.  For example, it may be the case that *quantum
physics* is a more specific topic than *consciousness*, but only in the
context of *philosophy of mind*.  In the context of *philosophy of
physics*--where
*quantum physics* is discussed in many contexts independently of *
consciousness*--the reverse may be true.

Most generally, the right model is one of overlapping semantic fields in
multi-dimensional semantic space.  More general topics have more "volume"
than more specific ones, and can contain those of terms they are broader
than.  This semantic space can be truncated by restricting one's attention
to a particular sub-region. For example, if one restricted one's search to
philosophy of mind, then we would chop off all areas of terms' semantic
spaces that fall outside that region.  This might change the containment
relations of terms remaining in that space.  Inside the semantic space
of *philosophy
of mind*, the space for *quantum physics* might be contained inside the
remaining space for *consciousness*, but this is only because most of the
region of *quantum physics* has been "chopped off" by restricting one's
attention.  I suspect that, in reality, most taxonomic semantic
relationships are context-sensitive in this way.

The question then remains how to best model such a context-sensitive
containment relation.  I see a few different approaches in SKOS, none of
which are ideal.  Perhaps most simply, one could treat each context of
interest as a separate conceptScheme.  But this would be rather ad hoc and
lacks generality.  Alternatively, one could try to create a specialized
suproperty of skos:broader, e.g. skos:broaderThanContextSensitive.  But
then one would need a way to independently specify the context in which the
relation held, and within which transitivites could be inferred.  This
would seem to be a metamodeling task, requiring a triple about another
triple.

Have others already considered similar complexities?  Any advice would be
greatly appreciated.  (Apologies if I've missed a relevant prior discussion
in the archive!)

Best Wishes,

Cameron Buckner
University of Houston
Department of Philosophy
713.743.2255

Alexander von Humboldt Postdoctoral Fellow
Institut für Philosophie II*
*Ruhr-Universität, Bochum

http://cameronbuckner.net/professional/