Re: How do you deprecate URIs? Re: OWL-DL and linked data

On Jul 9, 2008, at 1:20 AM, Richard Cyganiak wrote:

> On 9 Jul 2008, at 00:11, Bijan Parsia wrote:
> [big snip]
>> Complaining that the Big Nasty People Who Know What They're  
>> Talking About are raining on your sameAs parade isn't constructive.
>
> Ah Bijan. How about *you* grow up, flameboy?

Dude, did you flunk "Schoolyard Taunts 101"? I think the term you  
were looking for was "boogerhead".

Peter, however unintentionally, was advancing an "argument" that  
relied on disparaging a group of people. I'm in that group so it  
annoys me.

> You keep asserting that There Are Technical Problems With Using  
> sameAs. It would help your argument

I wasn't arguing it. I initial just raised the point.

> if you told us what those technical problems actually *are*.

No one asked, but here's two easy ones:

1) sameAs tends to merge annotations and similar meta data, because  
it's extensional.
E.g., If I say someTerm dc:creator "Bijan" and someone else  
someOtherTerm dc:creator "BoogerHead Jones", and then we say that  
someTerm sameAs someOtherTerm, we've (semantically) lost the  
distinguish between who created what.

2) In weak logics, sameAs are only inferred through explicit sameAs  
chains, so even if you do it between classes, it's pretty easy to  
resolve. In a sense, it's just the same thing as a renaming (which  
can cause problems, see 1). However, in stronger logics, like owl,  
where you can have inferred sameAs or *disjunctive* sameAs (i.e., A  
can be *either* equal to B *or* to C) things get very complicated,  
both from a user and from a reasoning perspective. If we try to  
preserve the "renaming" bit, then we have to move to Hilog semantics,  
which requires some restrictions to be decidable in the first place,  
and we don't have any experience implementing (even the decidable  
bits) in an OWL context.

Thus, it's not clear at all it will scale.

> I heard you say that using owl:sameAs could bite us in the butt.  
> Could you be more specific?

Sure, see above. I'd preferred if you had been politer about it.  
Please note that with Peter I was perfectly polite about the  
derogation in my first message. I only escalated because it was  
pretty clear that my complaint about being derogated wasn't coming  
through.

> Many people in this forum, including me, do not have a background  
> in formal logics.

That's reassuring to know about a designer of a KR language.

> Without that background, it is hard to distinguish proper uses of  
> owl:sameAs from improper uses of owl:sameAs.

If you use sameAs in just between individuals, then you will have no  
(extra) problems on the reasoning side. Please don't use it to align  
classes or properties. (Unless, of course, you are generally not  
planning on moving to expressivity involving substantive class  
subsumption inference.) If you care about annotations and the like,  
it's a good idea to avoid it. Obviously, if the alignment isn't  
precise (i.e., you think the terms are corresponding but not sure if  
there's an identity) you probably shouldn't use it either.

In modeling, it can be helpful to explore the consequences of making  
two individuals identical. But if you go on to publish that, I'd take  
care to reconcile any annotations or provenance information.

In general, I would encourage people to work on developing a more  
sophisticated alignment vocabulary.

> Please give us some guidance on that rather than wasting your  
> intellect on fanning the flames.

I'm curious to know why you think you weren't fanning the flames. I  
mean, it seems like you thought you were doing what I was doing. I'm  
curious to know why it's ok for you.

> A side note: The reason why I advocate the use of owl:sameAs is not  
> that it's the *right* solution. But it's *the only solution that  
> was available*.

This is, indeed, a problem. Let me be more precise: I see a lot of  
people advocating using sameAs as if it were not a stopgap, but a  
perfectly sensible, pitfalless solution. Those people (hi TimBL!)   
are doing a disservice, IMHO.

There are circumstances where it can work and work just fine. No  
annotations and no class-class, property-property sameAs (even  
indirect ones) are likely unproblematic.

> The alternative would have been to argue for a year or two instead  
> of linking up our datasets.

There are other solutions including some sort of transformation, or  
coining a mapping language.

> Not compelling.

Perhaps not.

> That being said, I'm very interested in hearing your take on when I  
> should use owl:sameAs and when not.

See above.

Cheers,
Bijan.

Received on Wednesday, 9 July 2008 07:13:35 UTC