Re: Issue-12 and the next UCR draft

On Sep 13, 2006, at 6:10 PM, Dave Reynolds wrote:

> Bijan Parsia wrote:
>> [snip]
>>>> There are already a large number of in-use or proposed semantic  
>>>> web rule languages (CWM, Euler,
>> These are not rule languages, but rule engines. The language they  
>> support is N3.
>
> They certainly both use the N3 syntax for rules, I wasn't sure if  
> they supported the same range of operators and builtins these days  
> and so whether they are really equivalent implementations of a  
> single language. Jos?

Well, pretty close, I'd say. Builtins in N3 are a bit like libraries  
or extended code, I think. But I take your point.

>>>> By phrasing this goal as "provide the basis for ..." we are  
>>>> indicating that there is unlikely be a single semantic web rule  
>>>> language and that RIF will not propose one. However, it also  
>>>> says that RIF should go further than minimal compatibility and  
>>>> try to bring some order to the chaos of semantic web rule  
>>>> languages.
>> Good luck :)
>
> Let me be clear on this. I'm not saying "the working group must do  
> this" I'm saying "the working must say clearly whether it is doing  
> this or not".

Oh, yes. That's a good idea ;) Let me change it to, "We probably want  
to say 'Not'", alas.

> Personally, about 2 hours into the first f2f meeting I suspected  
> the answer was probably going to end up as "no, we're not". I just  
> want clarity, partly for expectation setting, and partly (being  
> parochial here) because it will affect how HP regards my time being  
> spent in RIF.

Oh yeah. Manchester too!

>> There are deep issues with BNodes that have hurt us in SPARQL, and  
>> I think most extant SemWeb rule languages largely punt on them. It  
>> would be good to deal with them properly. (e.g., are BNodes scoped  
>> to the document? Even when they appear in rules?)
>
> Agreed there are dragons there.
>
> [quote from separate msg]
>> Sandro, the problem is that most extant rule languages do *not*  
>> treat BNodes as existentials, but skolemize them in a variety of  
>> ways. There are subtle and tricky issues esp. when talking with  
>> people with different views of how they work. Then, once you have  
>> agreement there, how they integrate with logic programming type  
>> rules is, again, very tricky. It's a research topic on its own and  
>> many of the extant rule languages have not attended to them (which  
>> is fine, except for poor standards bodies that have to harmonize  
>> with existing specifications).
>
> Definitely an issue, there are places where the skolemize approach  
> appears to be more pragmatically useful

I think it's almost always the right thing and that RDF should be  
changed.

> - even though unclean.

There's nothing unclean, just incompatible with a standardly ignored  
and unimplemented part of a different spec. Alas, a spec which has  
power over us.

> Of course by saying "basis for ..." we are giving ourselves scope  
> to not answer those questions. We could punt entirely on bNodes, we  
> could give partial treatments (no bNodes in heads) or whatever.
>
> Or we could decide that it is too hard or too early to do anything  
> along these lines and we say so clearly in the UCR rather than ask  
> for comments on the issue.

That sounds v. good.

Cheers,B
Bijan.

Received on Wednesday, 13 September 2006 17:16:21 UTC