Re: SKOS transitive hierarchical relations

On Thu, Jan 5, 2012 at 3:58 AM, Vincenzo Maltese <maltese@disi.unitn.it>wrote:

> Dear Simon, all,
> I'm surely not an expert of SKOS, nor a librarian, but at the latest UDC
> seminar I participated I tried to bring the point of view of my research
> group (working in knowledge representation and reasoning) in which we
> tried to formalize the meaning of the BT\NT relations.
>
> We interpret them as superset/subset and let them correspond to
> subsumption in description logic were the interpretation of each node in
> the vocabulary is the set of documents about the node. Now, since both
> subset and subsumption are transitive, this works only for transitive
> BT\NT relations. We treat non transitive BTs as associative relations.
>

This is mostly correct; associative relationships should be used where the
 relationship is not  necessarily transitive for the applicable domain of
discourse (which is either the subject area for which the thesaurus is
defined (ISO 2788), or the entire bibliographic universe (NISO Z39.19:2005).

Note that the key distinction between Knowledge Organization Systems and
Knowledge Representation systems is that the latter have extensions that
arbitrary things, whereas KOS's are extensional over Conceptual Works...
Word and Subject v. Word and Object.  Or to put it another way...
http://www.ibiblio.org/ses/anyqs.jpg

BTW, to clarify the Turbine and  Blade example;

Under the ISO rule, for a thesaurus specific to the domain of documents
about turbines, if all documents about blades are necessarily documents
about things which are parts of turbines, the vocabulary engineer may use
the BT relationship, since there cannot exist documents about blades which
are not about turbines.  This relationship is prohibited under NISO
Z39.19:2005.

Transitivity of NT does not entail any relationship between co-hyperonyms
of a given subject.
e.g.
(Turbine Blade NTP Turbine) AND (Turbine Blade NTG Blade) |= (Blade NT
Turbine)

(Parrots as Pets NT Parrots) AND (Parrots as Pets NT Pets) |= (Parrots NT
Pets) OR (Pets NT Parrots).

That most parrots in Western Europe and the US are pets is not expressible
in a monotonic, non-probabilistic   ontology language.  More expressive
languages can express this. PR-OWL (Kathy Lasky, Paulo Costas, and their
students) can conditionalize the probabilities (P(Parrot is Pet | Location
is Fairfax, VA) = 0.95); a system like Cyc can add the axiom as a default
within a microtheory specialized to urban areas (Parrots that are not
abnormal are  Pets).  Abduction can also allow one to hypothesize that a
Parrot is a Pet, but that situation is better handled through a
probabilistic reasoner.

Part of the problem with the way that SKOS handles broader^* is that the
current rationale is based on  epistemic/doxastic/illocutionary  concerns
that could arise when using a reasoner that cannot could not distinguish
between inferred and asserted object property values.  Since this
information is part of the OWL API (cf
OWlIndividual.getObjectPropertyValues<http://owlapi.sourceforge.net/javadoc/org/semanticweb/owlapi/model/OWLIndividual.html#getObjectPropertyValues(org.semanticweb.owlapi.model.OWLObjectPropertyExpression,
org.semanticweb.owlapi.model.OWLOntology)> with
OwlReasoner.getObjectPropertyValues<http://owlapi.sourceforge.net/javadoc/org/semanticweb/owlapi/reasoner/OWLReasoner.html#getObjectPropertyValues(org.semanticweb.owlapi.model.OWLNamedIndividual,
org.semanticweb.owlapi.model.OWLObjectPropertyExpression)> )this does not
seem to be a problem with the reasoners in current use.

If the  underlying problem  is how to handle vocabularies that use BT
incorrectly, better approaches are to either fix the vocabulary (if the
vocabulary is supposed to be using BT in a standard form), or, if that is
not possible, to translate the relations accidentally labeled as BT as
being associative.  This is the treatment for quasi and pseudo broader
relationships prescribed in the standards.

Simon
p.s.
  There are possible psychological reasons to doubt the cognitive validity
of the broader term relationship, but these reasons are explicitly blocked
in the Standard's definitions of controlled vocabularies (most questions
arise out of  polysemy and from prototype effects in psychological
concepts).