Hi all

SKOS concepts are likeky to somehow 'match' similar individuals in other KOS, and singularly OWL ontologies. This question has already been discussed with no clear answer on what should be a sound/recommended/best practice. I'm currently working on a typical use case which will maybe help to illustrate this difficult issue, namely geographical-administrative entities. The project involves the French National Institute for Statistics and Economic Studies (INSEE) [1], which is the official provider of facts and figures concerning the country and its administrative subdivisions. The project aims to define a RDF representation of the involved entities, unformally but precisely defined by INSEE in its 'Code Officiel Géographique' [2]. Administrative structure of France being quite complex and multi-hierarchical, capturing it in a KOS is quite a challenge, and seems to need all the expressive power of OWL. The ontology of those entities relies on a backbone of 'administrative subdivision' relationships, and a bunch of constraints over those, such as : A 'department' subdivision is an instance of 'arrondissement', with a 'chef-lieu' (instance of 'city') which is either the unique department 'prefecture' or a 'sous-prefecture' etc ... The ontology has also to be extensible to similar entities in other countries, supporting e.g. the European nomenclature of NUTS [3].
Such constraints are useful to control ontology integrity, but many 'light-semantic' applications, such as search engines, will need actually only a simplified view of this ontology, with thesaurus-like relationships between entities, used for semantic expansion of search. For such uses, a SKOS representation of geographical entities and their hierarchy would be good enough, and such a representation could be proposed as a 'simplified view' of the ontology.
So the question is : what should be the (recommended) practices to provide such a simplification? Two main options :
  1. Don't do that! different uses, different semantics, different represenations. Entities in the SKOS representation should be defined independently of the entities in the OWL ontology, with different URIs supporting different semantics. A city is not a concept, having an individual both of type skos:Concept and a:Geo-entity is not a good idea.
  2. Do it for semantic integration : same individual, one URI. The OWL representation and the SKOS representation will not be used by the same applications anyway, so there is no practical risk in having individuals being declared of type skos:Concept in thesaurus-like vocabularies (SKOS), and of type a:Geo-entity in ontology-like vocabularies (OWL). Having  a single URI would be useful in an integrated environment using both indexing and search of documents indexed on geo-entities, and semantic query and inference on these entities.
Option 1 is safer, but raises the issue of semantic integration. How will I assert that this SKOS concept and that OWL entity are somehow representing the 'same' individual, and what is the meaning of this 'same-ness'? I won't push again 'hubjects' here, although I could ;-) . Option 2 is my favorite those days, following the arguments pushed lately by Pat Hayes [4]. But what I wonder is to which extent the 'simple' SKOS classes and properties should be tied to the 'complex' OWL classes and properties, for instance should we
Such declarations could be useful for OWL-to-SKOS migration, but are likely, if included in the OWL framework, to bring unsuspected and weird entailments ...

Any ideas/suggestions on this are welcome.


[1] http://www.insee.fr/
[2] http://www.insee.fr/fr/nom_def_met/nomenclatures/cog/index.asp (in French)
[3] http://en.wikipedia.org/wiki/Nomenclature_of_Territorial_Units_for_Statistics
[4] http://lists.w3.org/Archives/Public/public-swbp-wg/2006Jan/0139.html


Bernard Vatant

Knowledge Engineering

3, cité Nollez 75018 Paris France

Tel. +33 (0) 871 488 459 

Mail: bernard.vatant@mondeca.com

Web: www.mondeca.com

Blog : universimmedia.blogspot.com