- From: Jeremy Carroll <jjc@hpl.hp.com>
- Date: Mon, 11 Feb 2008 09:05:44 +0000
- To: Jim Hendler <hendler@cs.rpi.edu>
- CC: "Web Ontology Language ((OWL)) Working Group WG" <public-owl-wg@w3.org>
I found Peter's clarification of the semantic options helpful; perhaps in allowing me to understand why I was uncomfortable with Jim's proposal. Jim seems to be suggesting something inbetween Peter's "D/ Informal" and his "C/ Operational". I fear we would find a long term loss in interoperability, for a short term gain in ease of reaching a design. I note that any of Peter's A/ Model theory E/ Transformational or F/ Axiomatic can give the same level of rigour and, since we start with model theoretic approaches we may as well continue with them, if we want this level of rigour. Jim appears to not want that. === I do want to maintain significant degree of rigour with the definition of OWL Full for a variety of reasons. 1) Architectural My understanding of the Semantic Web is that the semantics is attached to the RDF graph, and different layers can *add* additional semantics to a graph by means of providing semantic extensions (and/or definitions of classes and properties in terms of ontology languages such as RDFS and OWL). Since these semantics are additions within a strict monotonic paradigm, a certain degree of interoperation is provided. There are a variety of errors in the history of RDF prior to the introduction of the RDF Semantics (e.g. containers and reification), where an informal and partially operational semantics results in systems that do not have such well behaved semantics and have unpredictable behaviour. So the model theoretic semantics provides a discipline against error. 2) Minimize change I personally see OWL 1.1 as a small change on OWL 1.0. The two key changes I believe are QCRs and subproperty chains. I am less than happy with almost any change over and above these, other than minor corrections. 3) Product positioning I don't see HP's OWL Full offering as a "scruff's only" deal (in the old AI scruffs v neats debate). I see the choice of whether to use OWL (for both Full or DL) in a scruffy or neat fashion as an end-user decision. The choice of Full or DL is primarily a choice about expressivity vs computability guarantees. In typical OWL Full usage you get to use language features that you want, including ones that are not permitted in DL. You use these in the fashion that you want (whether neat or scruff). Then maybe an off-the-shelf reasoner and reasoner configuration gives you the trade-off you want between sufficient entailmenets vs speed - if not, you might need to customize your reasoning by selecting say, some rules out of a larger set of correct rules that cause a computational explosion. I fear that only having an informal, operational semantics for OWL Full would result in OWL Full being narrowed to only certain informal and less than precise work. Jeremy
Received on Monday, 11 February 2008 10:10:22 UTC