- From: Adrian Walker <adrianw@snet.net>
- Date: Thu, 20 May 2004 14:29:40 -0400
- To: Dan Brickley <danbri@w3.org>
- Cc: Richard Lennox <listserve@richardlennox.net>, www-rdf-interest@w3.org
Hi Dan -- You make a good point of course (see below), but it does not solve the problem The problem is that RDF/OWL is by design a machine-oriented notation, and one in which it is easy even for expert people to make mistakes when they try to code 'raw'. For example, a now apparently defunct web site (interprise.com) last year gave an example that can be summarized as some-subject is related by rdf:type to some-subclass that-subclass is related by rdfs:subClassOf to some-object -------------------------------------------------------------------------------------- that-subject is related by rdf:type to that-object The rule looks reasonable, and on the data that interprise supplied, its conclusions were in line with normal intuition** However, if you plug in "Clyde", "elephant" and "species", you get " Clyde is related by rdf:type to species" . So presumably, the rule is wrong as it stands. Yet, we need a rule of this nature, so what to do? It will not be enough to give a different rule at the machine-oriented level, because it also could be too easy to misunderstand. A possible improvement is to label each predicate with an English sentence for human use, and tie the labels firmly to the underlying machine notation. Consider*** some-item is a member of the set some-set that-set is a named subset of the set some-superset ---------------------------------------------------------------------------- that-item is a member of a named subset of that-superset The conclusion is careful to only tell a human reader that "Clyde is a member of a named subset of species". Of course, simple examples like Clyde-elephant-species do not seem all that important at first sight. However, if people can easily write a simple example to give an absurd conclusion, then we really do need to improve the situation for real-world tasks. The above is one way of trying to improve things. Perhaps there are others? Cheers, -- Adrian ** You can run this example, called RDFreasoning1, by pointing a browser to www.reengineeringllc.com and logging in to the "demo" ID *** Likewise for ClydeElephant1 INTERNET BUSINESS LOGIC www.reengineeringllc.com Dr. Adrian Walker Reengineering LLC PO Box 1412 Bristol CT 06011-1412 USA Phone: USA 860 583 9677 Cell: USA 860 830 2085 Fax: USA 860 314 1029 At 10:44 AM 5/20/04 -0400, you wrote: >* Adrian Walker <adrianw@snet.net> [2004-05-20 10:46-0400] > > Richard -- > > > > Actually, I think you may have put your finger on a general problem..... > > > > The general problem is, deductions that seem OK in RDF-ish notation are > > sometimes OK in English, and sometimes absurd. > > > > There's a nice example from John Sowa about this: Clyde is an elephant, > > elephant is a species, therefore Clyde is a species. That's wrong in > > English, but there are ways of writing it in RDF/OWL that look OK. > >This is a classic example, stemming from two senses of 'isa'. Modern KR >systems including RDF/OWL distinguish them carefully. In RDF, we have >'rdf:type' and 'rdfs:subClassOf' to represent two notions which >might colloqially be described as 'is a' by an English speaker. In RDF, >'type' relates an individual to a class; 'subClassOf' is a relation >between classes. > >Dan
Received on Thursday, 20 May 2004 14:43:44 UTC