Re: Well Behaved RDF - Taming Blank Nodes, etc.

On Dec 14, 2012, at 15:42 , David Booth wrote:

> On Thu, 2012-12-13 at 15:14 -0500, Ivan Herman wrote: 
> [ . . . ]
>> I think it would still be better to explain these things in a syntax
>> independent way. After all, I may want to use JSON-LD or RDFa...
>> Distilling the various mails and concentrating on bnodes only, what
>> seems to be the pattern is 
>> - bnodes can appear in at most one triple as an object
>> - there can be no cycle in the graphs involving bnodes
>> Would that suffice as a more formal definition?
> As of today I think that would suffice, though I'm unsure of the details
> of how the "no bnode cycles" constraint should be formalized.  Perhaps
> Jeremy Carroll or one of the authors of 
> could comment.  

I do not know about Jeremy... But isn't it enough to say:

'if (a,b,c,...,x) is a cycle in the graph, then at least one of the resources in the cycle must be a URI reference?'

By taking that node as a 'top' level in Turtle (or similar) the rest of the cycle can be put into a

A [
    p q r.
    u p A.

form or equivalent.

The interesting question is whether, for example, canonicalization becomes easier and, mainly, deterministic with such graphs. My hunch says yes, because it is probably possible to create a canonical turtle file (for each cycle, take the lexicographically minimal URIRef as the top level, etc). That may be an interesting point on these well behaved RDF-s


> But Turtle currently does not allow inverse property notation:
> If that feature were added and the syntax independent definition were
> still going to track what can be expressed in Turtle without explicit
> bnodes, then the definition would get more complex.
> -- 
> David Booth, Ph.D.
> Opinions expressed herein are those of the author and do not necessarily
> reflect those of his employer.

Ivan Herman, W3C Semantic Web Activity Lead
mobile: +31-641044153

Received on Friday, 14 December 2012 20:59:04 UTC