- From: John McClure <jmcclure@hypergrove.com>
- Date: Fri, 8 Jul 2005 10:59:36 -0700
- To: <semantic-web@w3.org>
Thanks all for the responses. 1. Danny Ayers points out correctly that the same semantic (that X is a "Person") can be inferred if the 'range' for "hasParent" is a "Person".... true, but here I'd like to address the implications of the UBL reqt that names of its elements (particularly, its property names) be nouns, NOT predicates. 2. Ian Davis suggests the alternative <Person> <parent> <Person rdf about='uri'/> </parent> </Person> for my <Person><has><Parent rdf about='uri'/></has></Person>. My concern is that <parent> is a noun, in effect, thereby eliminating predicate verbs as meaningful classes (for there is <willHave><Parent>, <had><Parent>, and other similar predicate verbs); that, as a property, it has not the expressiveness of restrictions, unions, and intersections available via owl:equivalentClass; and that it lacks the quality of being able to make statements about or establish membership within 'a set of (all) parents', that is, one cannot say that a (Person rdf:type Parent). Ian asks about <foaf:knows>.... while not a master of FOAF, I think the equivalent expression using predicate nouns would be <Person><has><Acquaintance rdf:about='uri'/></></>.... surely, one 'knows' one's acquaintances to some degree. 3. Frank Manola notes that some don't use <hasX> implying he has seen <has><X> before.... who would they be I am wondering. Other than rdf:type, *all* OWL property names confabulate the predicate verb with a predicate noun, consistent with fundamental RDF architecture, Fact=Subject+Predicate. This architecture clearly results in a "RDF Naming Convention". Frank also states that "the 'hasFoo' approach has the merit of indicating the directionality of the relationship" a feature which, I believe, is preserved by <has><X>. Like Frank, I also value readability highly, and therefore am cautious about designs which eliminate predicate verbs from markup altogether. Frank also suggests I take this question over to the SWBP list (see below). ------------------- Why is this issue more than just a syntactic game? My problem stems from the fact that domains (and ranges) can be specified only for properties, not for classes. These nouns indicate sets of Things; they are undoubtedly classes. But, how does one specify that for instance, only a LivingThing can have a Parent? One cannot with OWL DL, but of course in OWL Full, I can create metaclasses such as Noun (subClassOf Class), Verb and Preposition (subClassOf Property), and DirectObject (subClassOf Noun and subClassOfProperty). Then, define 'Parent' as an instance of a Noun (and subClassOf Role), and define 'CustodialParent' and 'BiologicalParent' as instances of DirectObject, with range (or subClassOf) 'Parent' and with domain (sic) Person. This yields what to me is a most sensible statement: <Person><has><CustodialParent rdf:about='uri'/></></>. In other words, the architecture is that the name of the direct object is *never* a type of object (Parent) but rather it is *always* its functional role with respect to the subject of the 'statement' being made. Here, CustodialParent reflects the role that a Parent plays with respect to the Person. Note also that CustodialParent can easily function anonymously.... <Person><CustodialParent/></Person> can imply the <has> predicate verb. Finally, note that the range for <has> is flaccid: it is just a DirectObject. So the conclusions I'm presently drawing are (1) it is not possible to create a UBL conforming ontology using OWL Lite or OWL DL (2) in OWL Full, a grammatical foundation is both feasible and immediately useful With regard to the suggestion that this be handled by SWBP, I would be fine doing so but am wondering also whether the matter that RDF has yet to define constructs like <Noun>, <Verb>, and <DirectObject> has traction here in semantic-web, since I'd think the reach of the SW would be considerably greater should UBL requirements be accommodated as soon as possible. *RDF Naming Conventions* is clearly a ripe issue for SWBP but your concurrence with (1) and (2) above would provide excellent guidance for that discussion. Thanks, John McClure PS I admit to being 'just' an XML guy, so please bear with me if I mangle any N3 expressions!
Received on Friday, 8 July 2005 17:59:30 UTC