- From: Jeremy Carroll <jjc@hpl.hp.com>
- Date: Wed, 22 Jan 2003 09:31:12 +0100
- To: www-webont-wg@w3.org
This is a follow up message to my proposed change to AS&S: http://lists.w3.org/Archives/Public/www-webont-wg/2003Jan/att-0356/01-jjc Overall, I believe that having the abstract syntax does allow a clear semantics to be given to RDF graphs, but I am concerned that every draft to date of the AS&S has had a concrete syntax (RDF graph for OWL DL) that has either: - only been described implicitly - or described in convoluted and difficult to follow text Moreover the concrete syntax for OWL Lite has never been articulated (except implciitly). I believe that these facts would present severe difficulties to both tool developers and OWL end users in actually using OWL Lite and OWL DL, since it is too difficultt: - for tool developers to understand what is and what is not a legal OWL Lite or OWL DL document - for document authors to understand the same - and for tool developers to create error messages that communicate to document authors would they need to do to fix a document intended to be OWL Lite or OWL DL. I believe that the approach at describing the concrete syntax developed in my message would better allow for all three of these. == This does involve two significant changes: - first a change in mentality, away from the abstract syntax as driving the show (because of its connection with the semantics), and towards a more balanced approach in which concrete syntactical concerns influence the abstract syntax (sections 2 and 4 of AS&S should reflect a trade-off between the semantic needs and the concrete syntactic needs) - second a few minor changes to the abstract syntax and mapping rules to make the proposed characterization correct. (a current list is detailed below) I am not proposing that my characterization be normative, (although I do point out that it is currently the only correct characterization of the side condition on TransitiveProperties). === "Minor" changes, in order of magnitude: - dataRange Under my proposal this either needs to be deleted entirely from the language (my preference) or assigned a type (e.g. owl:DataRange). - annotations Under my proposal the properties in annotations are owl:ObjectProperty's and owl:DatatypeProperty's and can be used elsewhere in an ontology) However they may not be used in domain and range constraints, in any restrictions, or in transitive, symmetric, functional or inversefunctional properties It is open whether we would want to restrict relationships between ontologies as only being owl:imports, owl:backwardsCompatibleWith, owl:incompatibleWith, owl:priorVersion, or whether we allow this set to be user extensible. It is also open whether such relationships may be used elsewhere in the ontology (as owl:ObjectProperty's without further constraints). - restrictions The abstarct syntax for restrictions needs to be changed to permit exacltly one restriction in each restriction construct. This makes no difference to the expressive power of language, since every use of the restriction construct can be repeated. - Equivalent descriptions The mapping rules are changed slightly to provide n-1 sameClassAs for n equivalent classes instead of the current n(n-1)/2. - DisjointClasses The abstract syntax is altered to express this pairwise, instead of permitting a list. (motivation this is the only place in the mapping rules in which a bnode becomes the object of more than triple - since this is an easy rule to articulate and understand, it is better to change the abstract syntax and not break the rule). drawback it is harder to express a list of disjointclasses (but then we decided to drop DisjointUnionOf from the concrete syntax - so I don't think we saw this as a big worry). - types The mapping rules need changes to give every node a type I think that's it - but I suspect there's a bit more. == I would be happy to work with Peter to integrate my text into the AS&S. Jeremy
Received on Wednesday, 22 January 2003 03:30:17 UTC