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

Re: Agenda request: characterize the diffs between subgraph-matching and E-entailment

From: Bijan Parsia <bparsia@isr.umd.edu>
Date: Mon, 9 Oct 2006 20:57:27 +0100
Message-Id: <A9B093C7-FCFD-4D2B-9E0D-8386DFE6695F@isr.umd.edu>
Cc: RDF Data Access Working Group <public-rdf-dawg@w3.org>
To: "Seaborne, Andy" <andy.seaborne@hp.com>

On Oct 9, 2006, at 8:24 PM, Seaborne, Andy wrote:

> Bijan Parsia wrote:
>> On Oct 9, 2006, at 5:55 PM, Seaborne, Andy wrote:
>>> Bijan Parsia wrote:
>>> . . .
>>>> I believe not if the bug is repaired a la Enrico's post.
>>> Which post are you referring to?  The last one in the archive  
>>> from Enrico is August 11.
>> Er...I don't know if there's a specific fix post. I just meant  
>> that there was a bug in that the two semantics didn't align, but  
>> that's a repairable thing, IIRC.
> That's the claim and I hope it's true.  I haven't seen anything yet  
> that gets us to revised text and that concerns me.  It will take  
> quite some discussion to stabilize the changes.

I think there are changes that have to be made in the presentation  
overall. One thing I died on is that the *rationale* for various bits  
has to be handled.

But one thing at a time, eh?

> What we do have is consensus that the implementation hint [1]  
> should match the revised semantic definition for the cases where  
> they overlap.


> > It doesn't have anything directly to do with
>>> . . .
>>>> *Evidently* we need to refresh people's understanding of the  
>>>> panoply of issues, and, frankly, the best way is to start good  
>>>> email threads.
>>> We do have test cases that have been around for a while.  We  
>>> could start by email about those.
>> Sounds good. But could we deal first with some of these algebra  
>> issues?
> Only if we take the implementation hint for SPARQL defined for  
> simple entailment (which is this round of standardization).


> Otherwise, building test cases isn't much use.

I guess I'm lost.

>   We can't write test cases on top of a changing BGP semantics  
> unless we rigidly assume the implementation hint.

Because the answers might change?

> Test cases are:
> 1/ A deliverable we are chartered to provide
> 2/ A good way of making sure we are talking about the same thing.

Sure, but I think progress on the semantics of the algebra are  
achievable even without test cases, or with tests cases that might  
have to be updated (just as they were with distinct).

Anyway, that's the action I had from last week. I'm fine starting up  
the discussion of the semantic framework, but 1) I have to get it  
swapped in and 2) clearly a lot of explanation has to happen. I get  
frustrated, I think understandably, when the framework is thrown up  
in a tangled, confused and confusing way, and I hear that we're going  
to tackle it when, and, oh yes, lets start with some aspects that,  
well, aren't aspects.

>> I say this because I only have so much energy and attention at the  
>> moment (aside from being overwhelmed with stuff, I'm having  
>> ongoing, we believe arthritis med related, complications which are  
>> draining). I'm a bit surprised to see how many actions I took on  
>> last week!
> Sorry to hear that - I hope any complications will be brief.

4 weeks. Counting. Sux. But in a different phase. New interesting  
effects! Always exciting.

>> Claudio and Jorge's papers and posts are publically available, so  
>> I'm totally not the gatekeeper.
>> Unless there is some reason to believe that the semantics of BGP  
>> and the algebra are tangled up more than we might have different  
>> inputs to the algebra...there's no need to couple or order them in  
>> any specific way, yes?
> I have always proposed keeping them separate as they are in my  
> mind.  And no ordering is therefore fine - we have test cases that  
> have been submitted about entailment and have had for a while now.

Well, test cases for BGP can just say what the expected answers are,  
of course. If our semantics doesn't support those answers they become  
negative test cases.

However, I'm well in favor of examples where people think the LC1 and  
LC2 semantics require different answers.

Just to kick off other considerations, I'll note that if we go to LC1  
style presentation, work needs to be done because, it turns out, that  
RDFS doesn't necessarily have unique (maximal) up to isomorphism  

See <http://www.sigmod.org/pods/proc04/pdf/P-10.pdf#search=% 

  See example 5.

(Note that there are several different notion of closure, as well as  
a notion of normal form. The working group, of course, *could* choose  
to punt on RDF, RDFS and various flavors of D, but they should then  
explicitly punt, instead of hinting that the semantic framework can  
handle them. See our discussion of contradictory graphs.)

Oh, 5.2, shouldn't we have the application of the framework in a blue  
box or something. I keep missing it.

I need to see the bug that causes misalignment again.

Received on Monday, 9 October 2006 19:57:43 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 15:00:52 UTC