- From: Uschold, Michael F <michael.f.uschold@boeing.com>
- Date: Tue, 13 Sep 2005 18:24:52 -0700
- To: <public-swbp-wg@w3.org>
For various reasons, I propose to replace the current discussion section with a new version below: Most of the content is the same, but I have used a different 'story' to include it. I also got rid of the most contentious statement. === This approach achieves the goal of letting the automated reasoner will do the work of building a taxonomy of different kinds of parts, that reflects the explicitly defined part hierarchy. This alleviates the need to make explicit statements like the following to declare that Headlight (or any other part) is a subclass of CarPart: Class(Headlight partial CarPart restriction(part:partOf someValuesFrom(ex2:Car))) Instead the reasoner infers this information, resulting in a more compact representation of the ontology. It has been argued that such an approach increases maintainability and modularity [R-NORM <http://www.w3.org/2001/sw/BestPractices/OEP/SimplePartWhole/index.html> ]. The inference works properly with the existential restrictions on the direct part properties as shown. To get a classifier to infer the above hierarchy using universal restrictions on the partOf_direct property in the first pattern, there would also have to be minimum cardinality restrictions on the property. This is one reason that existential restrictions are often chosen in these circumstances. [MFU comment: this seems odd; I would think that choosing between existential and universal should be based on what is most accurate and complete, or what is necessary to get the right inferencing, not whether or not cardinality constraints are necessary. And, what is the problem with having cardinality constraints? Is it a matter of convenience? Correctness? efficiency? Does the argument above about problems with cardinality constraints apply here?] Note that we have chosen this representation primarily to facilitate the use of inference to produce a part class taxonomy from an explicit representation of a part whole hierarchy. This approach seems harmless at first glance, but there are some caveats. What this representation is literally saying is that all engines are car parts, and that all car parts are parts of actual cars. Strictly, this is not true. Some engines are boat engines, and an engine for a 1969 Porsche 911E is generally considered a "car part" regardless of whether or not it is in a car (it may be for sale). The right choice depends on what is needed in a particular situation. To the extent that the following are true, this approach may be the right one to meet your needs: * there is a pressing need to have the kind of inferencing required * if you have a single local application, and the assumptions are clearly stated (even if strictly false). * if the scope of the application only includes cars and car parts, in which case boat engines are irrelevant. * you don't need to distinguish between being an actual part of an actual car, vs. or being a part intended for use in some [unspecified] car * if there is no need to interoperate, or if all interoperating applications make the same assumption. Conversely, it may be the wrong choice if the above things are not true. Of particular importance is the case when you wish for your ontology to be re-used for a wide variety of contexts and shared in a broader community. That is when making assumptions that are strictly false can create problems. === I chose to throw out the contentious and confusion-inducing statement: "Recall again, however, that the intent here is to create an intuitive model of what "whole cars" are made of, not a schema for concrete instances of these classes." Is there anything that you (ChrisW) intended by this statement that still needs to go back in? -- NB: this is also included (deeply buried) in a more detailed review of the note. I am sending a separate message so it is more likely to get looked at. ============================================ Mike Uschold Mathematics and Computing Technology -- Phantom Works; Tel: 425 865-3605 Fax: 425 865-2965 Building: 3307 Cube: 33D3 ============================================
Received on Wednesday, 14 September 2005 01:25:04 UTC