Re: [RDF-CONCEPTS] Skolemization

There is way too much analysis going on here. Skolemization is simpler than this, and y'all are making it into something wierdly complicated. 

First, it is ALWAYS semantically valid to replace any IRI with a blank node. So there is no need to "mark" an IRI as "legal to be made into a blank node". The point of genid is more like: don't think that this name has any particular significance *other* than being a replacement for a blank node. Here is an analogy from real life: the use of "John Doe" to identify someone without really identifying them, when reporting a legal case. This is a special name which can be used grammatically as a name but everyone knows is not *really* a name. It is easier to say "John Doe" than to keep saying something like "the un-named defendant". Similarly, a skolem IRI is something that is an IRI as far as the logical grammar is concerned, to keep things easy to process, but is publicly recognizable as a "fake" name, so that people know enough not to look up John Doe in the phone directory, or try to do an http access on the skolem IRI. (The analogy is not exact because "John Doe" gets re-used in different contexts, whereas skolem IRIs are globally unique, we hope. So a skolem genid IRI is more something like "John Doe #346218".)

If you were to go on using a skolem IRI as an IRI, that would not be an error. If you were to replace a skolem IRI with a bnode that also would not be an error. If you were to replace a non-skolem IRI with a bnode that would also not be an error, but it would usually lose information. What makes skolem IRIs slightly different is that doing this to them does not lose any information.

A skolem IRI is a perfectly valid IRI, at least inside RDF, and you can go on using it in the outside world. Just don't expect to be able to find out anything more about what it is supposed to refer to, or be able to use it as a link.  Probably best to not call it an *identifier*. Its like an opaque name that isnt in any list of names, can't be used to locate or identify anything, it just lets you know that there is something it names, and that's all you know about it. 

Make sense?

Pat


On Jun 17, 2013, at 6:28 AM, Markus Lanthaler wrote:

> On Monday, June 17, 2013 1:14 PM, Andy Seaborne wrote:
>> I'm not proposing an approach - I just outlined some different usages
>> that seem, to me, to be behind the discussion.
> 
> Yeah, and I just tried to understand that :-)
> 
> 
>> You picked === (identity equality, not equivalence) which I understood
>> to be that a claim that the IRI is identical to the bNode. 
> 
> That's what I meant.
> 
> 
>> IRIs do not
>> work like bNodes - we're agreed on that - that's at the model theory
>> level.  Skolemization looses something, we need to make sure it isn't a
>> harmful loss.
> 
> Agreed, I think at the moment I'm just trying to fully understand the
> consequences (on all levels) of skolemization because I'm not really sure I
> do understand how those well-known IRIs are supposed to be used or
> interpreted. At the moment I have the feeling that there are two
> interpretations of those well-known skolem IRIs. One is that they are
> "special" and more or less just a hack to work around a limitation in the
> RDF data model. Everyone seeing such a IRI will know that it actually is a
> blank node. The other is that they are just normal IRIs to the outside
> world, i.e., global valid identifiers. Once they leave my system, they are
> just like every other IRI.
> 
> 
> --
> Markus Lanthaler
> @markuslanthaler
> 
> 
> 

------------------------------------------------------------
IHMC                                     (850)434 8903 or (650)494 3973   
40 South Alcaniz St.           (850)202 4416   office
Pensacola                            (850)202 4440   fax
FL 32502                              (850)291 0667   mobile
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes

Received on Monday, 17 June 2013 17:02:17 UTC