I forwarded your yesterday's email to onto-std@ksl.stanford.edu, since this
was also included in my reply to Fritz, whose recipients I just overtook
(well, we should install an intelligent recipient-management system ...).
Here some quick remarks concerning your points on nonunary inheritance.
1. We can distinguish between hierarchies over the argument positions of
nonunary predicates (as in your 'universal' Married relation), hierarchies
over nonunary predicates with fixed argument sorts (see example below),
and the combination of both. Varying your example, let us consider this
little 'couple' hierarchy:
couple(human,human)
concubinage(human,human)
universally-married(human,human)
regular-married(human,human)
honeymoon-married(human,human)
after-honeymoon-married(human,human)
same-sex-married(human,human)
Obviously, you can inherit useful information for, say, the honeymoon-married
relation from its super-relations, regular-married and above.
2. Perhaps we can treat some of the problems you mention by introducing a
canonical order over symmetric relation arguments, as by permitting only
regular-married(man,lady) or, better, regular-married(husband:man,wife:lady).
We can now instantiate universally-married(uhusband:human,uwife:human) to
either regular-married(uhusband:man,uwife:lady) or to the first alternative
same-sex-married(uhusband:man,uwife:man) or to the second alternative
same-sex-married(uhusband:lady,uwife:lady), where uhusband/uwife denote
'universal' marriage roles. Introducing roles instead of fixed argument
positions corresponds to proceeding from Herbrand terms to Ait-Kaci's psi-
terms. His work on inheritance between psi-term-described concepts seems
to be quite relevant for our current efforts in ontology.
3. Concerning your original problem formulation, one might ask whether the
relational sort hierarchy (the 'taxonomy') or some additional integrity
constraints (the 'axioms') should be responsible to keep things straight.
If you permit the full power of FOPC (or more) for the axioms, you can even
control the use of very irregular inheritance schemes. I think the KL-ONE
experience has shown that one should carefully balance the taxonomic and
axiomatic part of ontologies.
Cheers, Harold.
PS: I very much like your interpretation of the "black piano" example:
A theory of instruments should separate sound-relevant and -irrelevant
properties. The example was inspired by Nicola Guarino's "red apple"
example (GCG94), now with *two* sortal predicates.
(GCG94)
%0 Book Section
%A Guarino, N.
%A Carrara, M.
%A Giaretta, P.
%D 1994
%T An Ontology of Meta-Level Categories
%B Principles of Knowledge Representation and Reasoning: Proceedings of the
Fourth International Conference (KR94)
%E J., Doyle
%E Sandewall, E.
%E Torasso, P.
%I Morgan Kaufmann
%C San Mateo, CA
%P 270-280