W3C home > Mailing lists > Public > public-rdf-dawg@w3.org > July to September 2006

The SameAs bit (was Re: Summary of BNode redundancy options (at the moment))

From: Bijan Parsia <bparsia@cs.man.ac.uk>
Date: Fri, 18 Aug 2006 07:23:52 +0100
Message-Id: <60E39376-4881-4F6F-BD86-A7DA5B56E0C3@cs.man.ac.uk>
Cc: RDF Data Access Working Group <public-rdf-dawg@w3.org>
To: Pat Hayes <phayes@ihmc.us>

Pat, Please don't strip all the citations. While it's just two of us,  
and we're in the moment, it's not hard to keep track, almost, but  
it's bad for the record, i.e., for people not involved.

On Aug 17, 2006, at 10:29 AM, Pat Hayes wrote:
[restoring citations/quote key]
> [Bijan Parsia wrote:
>> [Pat Hayes wrote:

>>> A question to clarify intuitions, although it is not directly  
>>> applicable to SPARQL. Consider the following:
>>>
>>> :Bill :p :a
>>> :Joe :p :a
>>> :Bill owl:sameAs :Joe
>>>
>>> SELECT DISTINCT ?x [?x :p :a]
>>>
>>> If we are using OWL-DL entailment, should the answer set have two  
>>> bindings or one?
>>
>> Good question and not one that I've seen asked or dealt with. But  
>> I think there's a mild UNA wrt answer set redundancy implicit in  
>> most treatments of query answering against DLs. Or perhaps, no one  
>> has thought hard about distinct.
>>
>> Thinking about it, and how we handle these things in general, I'd  
>> prefer 1 answer, but of the following form:
>>
>> 	?x
>> 	:Bill = :Joe (or :Bill/:Joe, or some such).
>>
>> But I wouldn't want two answers. Because there aren't two *answers*.
>
> Ah, I'm glad I asked the question, this is what I thought you might  
> say. Seems obvious to me that that there are two *answers* here.  
> That they co-refer is interesting, but it doesn't make them into  
> the same *answer*. An answer is a binding to a variable, not what  
> that binding denotes. Clearly our intuitions differ on this.
[snip]

Well, it's partly intuition and partly extrapolating from my Swoop  
building experience. When reasoning is on, it's good to have all the  
terms that are equivalent inlined in the class tree. So, if A iff B  
and B sub C then the class tree looks roughly like:

	C
		[A, B]

We don't currently do that for individuals, but individual support  
was rather late and illthought out in general. There's also the  
problem of scaling it in a UI.

There are lots of questions to answer before we could determine a  
reasonable semantics for DISTINCT in the presence of the OWA, no UNA,  
and equality (see the trouble we have with just OWA and no UNA). I  
don't know of any work about it. But I don't find your intuition  
obvious. Consider the following kb:

	:bijan flyingTo :paris
	:parsia flyingTo :paris.
	bijan sameas :bijanParsia.

There are *two* answer to the question "Who's flying to Paris?" If  
you asked someone, say kendall, who was coming to a F2F and he said  
"bijan" and "parsia" i.e., two answers (and you knew that I'm bijan  
parsia and some people call me by my last name), I think you would be  
puzzled by kendall's response. In fact, I think you would treat it as  
one answer repeated with a different phrasing.

This isn't to suggest that this would be determinative of what the  
best behavior to support would be.

In a database context, an answer is a binding to a variable and an  
answer is a binding to the "denotation" don't give interestingly  
divergence answers. But we're out of them pleasant fields.

(It's not really to the denotation, right? I mean, we don't *know*  
what the actual denotation is, we just know that in every  
model :bijan and :bijanParsia denote the same thing.)

Now, I've talked to people who think that using identifiers sorta as  
labels is not a good idea. There's a sense that this is what's  
happening in these examples (since it would disappear with the  
UNA. :bijan sameAs :bijanParsia would be inconsistent).

Cheers,
Bijan.
Received on Friday, 18 August 2006 06:24:25 GMT

This archive was generated by hypermail 2.3.1 : Tuesday, 26 March 2013 16:15:27 GMT