OWL mismatch example: Taxonomies VS. the semantic web?

This is a follow-up to my previous posting with an example as requested by several people on the list (see the bottom of this email for the original posting).

 
Let’s look at a OWL Lite wines sample extract of grape, winegrape and CabernetSauvignonGrape for the semantic web (taken directly from section 3.1 of the the W3C OWL guide at http://www.w3.org/TR/owl-guide/) :
 <owl:Class rdf:ID="Grape"/>
 <owl:Class rdf:ID="WineGrape">
   <rdfs:subClassOf rdf:resource="&food;Grape" />
 </owl:Class>
 <WineGrape rdf:ID="CabernetSauvignonGrape" />
Note that Grape is a class, WineGrape is a subclass and CabernetSauvignonGrape is an instance (see the owlguide for details of how/why).
 
Here is the same OWL Lite sample rewritten as a taxonomy writer (*) may write it, except for descriptions, labels, information urls and other interesting stuff that has no semantic meaning:
 <owl:Class rdf:ID="Grape"/>
 <owl:Class rdf:ID="WineGrape">
   <rdfs:subClassOf rdf:resource="&food;Grape" />
 </owl:Class>
 <owl:Class rdf:ID="CabernetSauvignonGrape ">
   <rdfs:subClassOf rdf:resource="&food; CabernetSauvignonGrape" />
 </owl:Class>
 < CabernetSauvignonGrape rdf:ID="SomeSpecificArticleAboutCabernetSauvignonGrape" />
 
(*) I here refer to a typical taxonomy writer as a person that:
1) Wants to structure/classify “information objects” (articles, books, websites) according a taxonomy, so that users can easily find the relevant “information objects” for each topic (=class).
2) Currently has no additional requirements, obligations, interests or knowledge of semantics, descriptive logics or other formal aspects of the future semantic web.
3) Is interested in using the OWL Lite language, OWL editors, and OWL reasoners in order to save costs and possibly for standardization as well (as long as it does not increase costs).
 
In the sample, note that CabernetSauvignonGrape is now a class (instead of an instance) and the addition of a sample “information object” instance. This change facilities use of commonly available OWL Lite reasoners to answer all relevant classification questions, including “which articles exists that is about WineGrape ?” (answer: SomeSpecificArticleAboutCabernetSauvignonGrape).
 
The difference:
When comparing the examples, note in particular the different view on what is classes and what is instances + different semantics of the subClass construction.
 
Still not clear is exactly how grave the situation will be for semantic web software but I foresee considerable confusion. 
 
Comments ? 
 
/M. Christensen

++++++ ORGINAL POSTING FOLLOWS ++++ 

W3C has recognized the importance of taxonomies and is pushing “OWL Lite” as the new preferred way to express (web) taxonomies.  However, there appears to be a subtle but important semantic difference in how OWL is best used for taxonomies and how OWL is intended to be used for the semantic web. A difference that may destroy the semantic interoperability of OWL taxonomies and OWL for the semantic web.
When used for the semantic web the purpose is machine interpretability of Web content. Individual classes and instances describe the (metaphysical) things of existence, like the colour blue or the day of the week “Monday”. OWL mark-up for a document (or other type of information objects) are formal semantic representations of what is described informally in the documents. 
When used to express classification systems (classification hierarchies / taxonomies) the purpose is to structure information objects in libraries, databases, portals etc. Individual classes describe topics/subjects such as Marketing,  R&D, Sales etc. Instances may (*) be the information objects under classification! OWL mark-up for a document (or other types of information objects) describe what the document is about (i.e. its classification).
(*) The recommendation to represent classification of “information objects” as class instances is arguably a misuse of the IS-A relation but it offers many benefits for taxonomy minded users/developers. 1) A simple information structure very similar to a taxonomy. 2) Out-of-the-box OWL reasoner support for relations between topics/subjects (e.x. reasoner will know that a book manually specified to be about a specific subject is also about the broader (parent) subjects in the taxonomy). 1+2) Lower costs, faster, easier, less training, less development.
In summary, key OWL constructs such as instances and subclassing will be used differently depending on the intended usage. I predict that the difference uses of OWL may inhibit the semantic interoperability of OWL taxonomies and OWL for the semantic web. (Will mostly be a problem of a future semantic web because the web may lose a valuable backbone of structured information).
 

------------
Yahoo! Mail - Gratis: 6 MB lagerplads, spamfilter og virusscan