- From: Graham Klyne <gk@ninebynine.org>
- Date: Tue, 30 Nov 2004 08:30:53 +0000
- To: Dan Brickley <danbri@w3.org>, Tim Berners-Lee <timbl@w3.org>
- Cc: Stuart Williams <skw@hp.com>, www-tag@w3.org, Joshua Allen <joshuaa@microsoft.com>
At 21:06 29/11/04 -0500, Dan Brickley wrote: >* Tim Berners-Lee <timbl@w3.org> [2004-11-29 20:53-0500] > > > Yes, there were constraints on what you can do with literals from > > reasoning in OWL, just as there are constraints that you can't for > > example have a datatype property be an inverse functional property > > (like social security number is supposed to be). But I don't believe > > it is the place of the RDF layer to try to enforce these things. > >Aside: this constraint (no datatype IFPs) is only in OWL Lite and >OWL DL, I believe. OWL Full doesn't care, nor (as you say) does RDF/S. >FWIW I describe foaf:mbox_sha1sum as an owl:InverseFunctionalProperty and the >sky hasn't yet fallen in. I get complaints from DL-oriented implementors, >but afaik the usage is 'legal'. [I'm not sure this discussion belongs on www-tag, but I'm not sure where else to divert it] The last time datatype values came up in discussion, I seem to recall it was to do with literals as subjects... I've recently been doing some work on DL reasoner implementation, and I've noticed that for some of the more advanced flavours of DL (roughly, corresponding to OWL-DL, I think), the published algorithms do rather depend on having the datatype values separate from the other values (the term "concrete domains" appears in the literature), and further there is no provision in this work for these values to appear as the predecessor in an expression involving a role (the subject of a statement in RDF-speak). I've no idea if this is a fundamental limitation or just a consequence of the current state of the work that has been done to develop DL reasoning algorithms. AFAICT, there is no problem here with datatype IFPs. There's a paper by Jeff Pan and Ian Horrocks describing a DL tableau algorithm that deals with datatype values and qualified number restrictions but not inverse roles (SHOQ(Dn)) [1]. (This paper has a definite eye to DLs for the Semantic Web.) #g -- [1] http://www.cs.man.ac.uk/~horrocks/Publications/download/2002/Pan-Horrocks-datatype-2002.pdf ------------ Graham Klyne For email: http://www.ninebynine.org/#Contact
Received on Tuesday, 30 November 2004 09:03:59 UTC