Re: Two profiles: technical definition

On Thu, 2024-05-02 at 11:32 +0000, Franconi Enrico wrote:
> On 2 May 2024, at 12:41, Olaf Hartig <olaf.hartig@liu.se> wrote:
> 
> > 1/ Your grammar for the abstract syntax in both of the profiles
> > contains the restriction that triple terms cannot be nested. For
> > instance, the following 3-tuple (having another 3-tuple as its
> > third element which, in turn, has yet another 3-tuple as its third
> > element) would not be an RDF triple that one may have in an RDF
> > graph.
> > 
> > (:r1, rdf:reifies, (:r2, rdf:reifies, (:s,:p,:o) ) )
> > 
> > I am not against this restriction. I just want to call it out
> > explicitly.
> 
> I noticed that myself this morning. I would be in favour of
> arbitrarily nested triple terms, since (a) they make sense, (b) they
> are well defined in the semantics, and (c) they wouldn’t add any
> serious additional complexity in the algorithms (it is still pure
> matching, albeit in a structure with no predefined depth).

Fine with me as well.

> > 2/ The definition of the notion of a 'model' in the "functional
> > opaque" profile contains an additional formula within which it says
> > IEXT(IS(rdf:edge)). This assumes that IS(rdf:edge) ∈ IP. Would it
> > be useful to make this assumption explicit in point 3 of the
> > definition of the notion of an 'RDF simple interpretation'?
> 
> It would be redundant, since that condition plays a role only when
> rdf:edge is actually used as a property in the graph, in which case
> IS(rdf:edge) ∈ IP has to hold.

Okay.

> > 3/ Based on the definition of the "functional opaque" profile, it
> > is still possible to have two different triple terms t1 and t2
> > (i.e., t1 != t2) that an RDF simple interpretation maps to the same
> > resource, IL(SRE(t1) = IL(SRE(t2). Is that intentional?
> > 
> > (I understand that the literals l1 = SRE(t1) and l2 = SRE(t2) are
> > different---because SRE is bijective---but IL is not bijective, and
> > shouldn't be.)
> 
> I assume that this could be fixed by saying that this literal belongs
> to a special ad-hoc datatype, where the interpretations of different
> literals would be different.

Given this response, I assume the answer to my question is: no, it was
not intentional.

I agree that such a fix is possible. A more formal way of capturing
this fix would be to introduce a special subset of the set called
'literal', make this subset the co-domain of the SRE mapping, and
extend the definition of the IL mapping such that the restriction of
this mapping to the subset is bijective.

-Olaf


> cheers
> —e.
> 
> 
> > On Thu, 2024-05-02 at 07:05 +0000, Franconi Enrico wrote:
> > > 
> > > (I repeat a previous email, which could have been lost within a
> > > previous thread)
> > > 
> > > In order to make the upcoming discussion more concrete and
> > > technical,
> > > I have written down the formal definition of two profiles in the
> > > wiki:
> > > RDF-star profile “transparent” (namely many-to-many transparent)
> > > RDF-star profile "functional opaque” (namely many-to-one opaque)
> > > They rely on two distinct properties - rdf:reifies and rdf:edge
> > > (temporary name) - and on two distinct syntactic categories -
> > > tripleTerm and opaqueTripleTerm.
> > > Technically, they could be just merged into a unique profile,
> > > which
> > > actually could be RDF-star itself.
> > > 
> > > —e.

Received on Thursday, 2 May 2024 12:18:25 UTC