- From: Hugh Glaser <hg@ecs.soton.ac.uk>
- Date: Wed, 13 Jun 2007 10:27:21 +0100
- To: <semantic-web@w3.org>
I have a sense of full circle here. The short answer to the (intended?) original question is: A URI is always a URL. For the Semantic Web, any URI should resolve to a document with some RDF about it (as Tim would say, I think). There should be no non-resolvable URIs. I can publish one or more URIs that I think of as some object I know about, such as me. They are not the object or me, but because I publish RDF about them, others can decide whether they might be the same thing (synonym) as the URI they are publishing (with their own RDF, of course). It would be meaningless for me to use a URI without saying something about it, as no-one would have any idea what it was about (symbol grounding, private language). Of course, the Semantic Web is still embedded in the Web, using the excellent infrastructure it provides, and so there needs to be some backwards compatibility. Thus we find ourselves concerned about URIs that appear to be in both worlds. There are interim (long term?) solutions proposed to work around this, such as the 303 shenanigans. I would suggest that probably the best solution, had the Semantic Web come first, would be that such URIs would simply be string resources in RDF, just like my house address or telephone number, to be interpreted (rendered) by whatever means possible, in the case of a web page, a browser would know what to do with a suitable predicate/object. Although the Semantic Web did not come first, we should not lose sight of where we might go in an ideal world. As well as publishing the RDF about my URIs, I can publish, at the same place or elsewhere, RDF that indicates that I know of other URIs that I think are the same as my URI, in some sense. Such RDF will need to be used in context: as some have commented, sometimes URI1 will be considered equivalent to URI2, other times it will not, depending on the application (the reason the question is being asked), trust, provenance and perhaps how much the questioner is willing to pay for the service. Bernard's cities example is very relevant here. My preference is to separate out these concerns, and publish the equivalence information elsewhere. I do not like embedding such links in the substantive RDF. All this is knowledge about the URI, and so should be represented in RDF. An ontology is required to describe this, and as others have noted, owl:sameAs can be problematic. owl:sameAs is a bit different, and should possibly only be used between URIs that I own; it is also pointwise, which is not very scalable. Note that a naming authority is never the solution. It cannot represent the complexity of contextual equivalence. Also, What happens when naming authorities collide (NSF meets EU Framework)? And even worse, when you discover that the chemical sample you thought was new was the same as an existing one, after asserting all sorts of knowledge about it. We confronted a lot of this in building CSAKTiveSpace ( http://www.aktors.org/technologies/csaktivespace/ ) and subsequently built a service we called CRS (Consistent Reference Service) in its various incarnations. Basically, a CRS is a service for putting URIs into bundles. This is what we are now using for the rkbexplorer ( http://resist.ecs.soton.ac.uk/explorer ), although we are busy re-minting the URIs for this, and the curation of the co-referees is still an old version. The system uses CRSes, when appropriate, to decide whether to consider that two URIs should be considered the same in each context. So, for example, the rkbexplorer application considers a report and a journal paper with the same content might be the same, as we are only concerned with who works with whom on what. If we were concerned with authority, we might need to ensure that citations of the report were not conflated with the citations of the journal paper. The system is designed to use multiple triplestores with multiple CRSes, which can be provided from elsewhere (although this has not happened yet). And of course, the CRS then must link out to dbpedia, etc, providing the glue for the linked data. It seems that what the Linking Data Project is now trying to do is very similar to this. Will be great when we get this all working together. Note that this allows me to assert that I think that Tim1 is Tim2, even if Tim1's RDF or CRS disagrees. This is to be expected. We all have our views of the world, and I might just be right. Caveat emptor, if you come to one of my CRSes. How to discover where a URI is used in some triplestore/CRS? Ah, that is another story... A report on CRS is available at http://eprints.ecs.soton.ac.uk/11539/ . Best Hugh -- Hugh Glaser, Reader Dependable Systems & Software Engineering School of Electronics and Computer Science, University of Southampton, Southampton SO17 1BJ Work: +44 (0)23 8059 3670, Fax: +44 (0)23 8059 3045 Mobile: +44 (0)78 9422 3822, Home: +44 (0)23 8061 5652 http://www.ecs.soton.ac.uk/~hg/
Received on Wednesday, 13 June 2007 09:31:52 UTC