Re: Go ahead with pub

On Oct 2, 2006, at 11:06 AM, Seaborne, Andy wrote:

> Bijan Parsia wrote:
>> Sorry to be late, but there was some confusion about what was  
>> going  on and then I hit the class part of the week.
>> I think the status part should point to the issues list.
>> """Costs: Tableau-based reasoners (at least, the Pellet Demo  
>> example  7) rely on the current, more expressive semantics to  
>> match  implications that are not in a materializable RDF graph."""
>> No. Pellet uses BNodes as syntax for non-distinguished variables,  
>> as  that's what we were told was the likely syntax for non- 
>> distinguished  variables in SPARQL/DL. The semantics of *all*  
>> variables in SPARQL/ RDF is semi-distinguished.
>
> I'd be interested in tracing this back sometime.  Who told you/Pellet

Enrico. I think it was personal communication. Since it was about a  
future design, it's a bit moot, but you can kinda see how it falls  
out of the current semantic framework. (Though, weirdly, enrico  
seemed to forget this at one point :)). The scoping set  determines  
the binding of query variables *only*. BNodes are, naturally,  
existential without restriction, so even under a more expressive  
entailment such as OWL DL, they'll range over arbitrary entities. So,  
if we remove BNodes from the scoping set (which Enrico and Sergio  
have definitely be presuming, though I have no idea if/where they've  
said that in our archives), query variables will only range over  
names and be distinguished (in the head) or projected- 
away-"distinguished". This is different from the standard behavior of  
query variables in DL systems where not appearing in the head  
*always* goes with ranging over arbitrary entities. However, even  
with this change to the scoping set, BNodes still range over  
arbitrary elements, and, in fact, cannot appear in the head, thus are  
non-distinguished.

So, with no BNodes in the scoping set, we end up with three sorts of  
variables:
	1) Distinguished (binding to names + in the head; syntactically,  
these are query variables)
	2) Projected away distinguished (binding to names + in the body;  
syntactically, these are query variablers)
	3) Non-distinguished (binding to anything + only in the body;  
syntactically bnodes)

Since the scoping set is not sensitive to position of the variable,  
we can't use it to get the distinguished-non-distinguished behavior  
of query variables as in current implementations (and theory). We can  
only get all distinguished (with some projected-away) or semi- 
distinguished + non-distinguished. All implementations can easily  
adapt to the first circumstances (indeed, only pellet has non- 
distinguished variables, so it's pretty easy). The cost is of a  
significant bit of functionality. (How *useful* that functionality is  
open to debate.) No implementation implements semi-distinguished  
variables and as we've found out, it's unclear what the semantics of  
semi-distinguished variables should be in the DL case (e.g., what are  
the answers? what are *distinct* answers?).

We could allow for an additional scoping set for query variables in  
the body (but then we loose projected away distinguished variables  
when we allow for non-distinguished variables --- perhaps acceptable,  
though it would block making some queries much more efficient).

> and does  the working group have a record of it?

I hope this message suffices as a hook from now on. I did a few quick  
searches to find some earlier discussions but, well, it looks hard ;)

> I'd find it useful in bring some of the themes together as there is  
> a lot of material scattered over the email archive.

And the minutes and off line email :(

[snip]
>> I would like that to be deleted because it's confusing on several   
>> levels (e.g., why doesn't it apply in optional?) I don't  
>> particularly  care that it'd done before pub, but it seems an easy  
>> enough move. I  mean, it doesn't *change* anything!
>
> This text does seem out of date: the objection from DaveB was  
> withdrawn and I can't find a record of any other (I just checked  
> with Steve).
>
> http://www.w3.org/2001/sw/DataAccess/issues#disjunction
>
> It is confusing.  But I'd like the chair's permission as confirmation.

Oh Chair, please give permission!

Cheers,
Bijan.

Received on Monday, 2 October 2006 11:57:10 UTC