Re: Weakness of RDF? (was: Tuple Store, Artificial Science, Cognitive Science and RDF (Re: What is a Knowledge Graph? CORRECTION))

> On Jun 26, 2019, at 1:40 AM, Graham Klyne <gk@ninebynine.org> wrote:

Hi Graham
> 
> On 26/06/2019 07:58, Patrick J Hayes wrote:
>> [...] For AI purposes, RDF is absurdly weak and
>> inexpressive. But AI is not what it is trying to do.
>> 
> 
> I'm reminded of a most interesting poster by yourself and Peter Patel-Schneider, presented at ISWC 2013.  One of the results, as I recall, was that RDF semantics is so weak that any RDF expression can be satisfied by an interpretation with no more than 3 members in its domain of discourse (subject to absence of certain semantic extensions such as some of those in OWL).
> 
> It was only afterwards that it occurred to me:  this isn't a bug, it's a feature!

It's certainly not a feature. Its not really a bug, either :-)
> 
> As I see it, one of the key consequences of the RDF semantics is:
> 
> Merging lemma. The merge of a set S of RDF graphs is entailed by S, and entails every member of S.
> -- https://www.w3.org/TR/2002/WD-rdf-mt-20020429/#entail
> 
> (Which I don't see mentioned in the more recent RDF semantics spec, but I assume it still holds.)

In RDF 1.1, entailment is defined only between graphs, not sets of graphs. So you have to decide how to make your set into a single graph. See the technical notes in sections 5.1 and 5.2. But yes, the result still holds, in effect. 

> 
> My take is that this is what validates combining (i.e. merging) RDF from independent sources, which I see as one of the key advantages of RDF compared with popular data models that don't have an associated formal semantics - we have a rule for combining data that comes with an (admittedly weak) semantic guarantee.

Indeed, but this property of what one might call universal combinability is trivial for almost all formal logics (it is just the rule that A and B together entail (A&B) ). By itself it does not make the logic very expressive. And it hasn’t got anything to do with the extreme trivialization lemma that PFPS and I were talking about. 
> 
> Yet, the weakness of these semantics suggests to me that the formal semantics is making a minimum of assumptions about how the RDF is being used, hence less likely to "get in the way" of desired application semantics.

But this line of reasoning would suggest that RDF have no semantics at all, and just be a handy notation (as JSON-LD is described by its creators, where the lack of a specified semantics is treated as a feature.) Perhaps that would have been the best way to proceed, in fact, since even the minimalist semantics that RDF has is often ignored by developers and users. In which case this enrire discussion would be pointless, since RDF would then not have any clear relationship to any logic or inference scheme or to AI-KR; and I, myself, for one, would not have any interest in it nor any reason to have become involved with it. 

Pat

> 
> #g
> --
> 
>

Received on Wednesday, 26 June 2019 09:29:33 UTC