Re: RDF semantics redux

Dan Connolly wrote:
> On Tue, 2005-11-08 at 01:17 -0600, Pat Hayes wrote:
> [...]
> 
>>With the local option, we can define an answer as a substitution S such that
>>
>>G entails Q[S]
>>
>>where 'entails' is a parameter. Putting [entails] = [simply entails] 
>>gives us the current design,
> 
> 
> Really? My understanding is that the last call design does _not_
> work this way.

In fact, it provides sensible answers *only* if all the bnodes are
skolemised (i.e. PH's fake IRIs), and you disallow true bnodes (i.e. non
skolemised) in pattern solutions.

Example:

Given the graph:

{:Arthur a Person
:Arthur :siblings _:l34
_:l34 type collection
_:l34 first _:P55
_:l34 rest _:l35
_:l35 first :Susan
_:l35 rest _:l36
_:l36 first :Bill
_:l36 rest nil
_:P55 a Person
_:P55 :gender male}

And the query:

:Arthur :siblings ?L
?L first ?V

This {L/_:l35,V/_:l34} is a solution; moreover, there are *always*
infinite solutions (just substitute arbitrary bnodes for L and V,
they're all solutions). So, to get something meaningful you have to
minimise every answer set. In addition, I don't see any reasonable
minimisation algorithm which would give you the "told bnodes" redundancy.

Bottom line, to make the local option working you need to take
skolemisation seriously, which means that you avoid bnodes in datasets
and solutions (you use the "fake IRIs" instead).

True bnodes would be allowed in queries, to represent existential
variables (i.e. non distinguished).

Enrico and I had the impression that the WG disliked the skolemisation
business, so we designed an alternative semantics which avoids
skolemisation at all.

I'd like to point out that the "fake IRIs" solution doesn't take away
the need for de-skolemise, since you need a mechanism to detect "fake
IRIs" as they represent original bnodes.

--sergio

Received on Wednesday, 9 November 2005 09:07:47 UTC