Re: ISSUE-148: RDF Concepts - IRIs do *not* always denote the same resource

On 15 December 2013 23:30, Markus Lanthaler <markus.lanthaler@gmx.net> wrote:
> On Saturday, December 14, 2013 7:54 PM, Pat Hayes wrote:
>> Possible wording compromise....
>>
>> On Dec 14, 2013, at 8:26 AM, David Booth <david@dbooth.org> wrote:
>> > On 12/14/2013 07:04 AM, Markus Lanthaler wrote:
>> >>
>> >>   . IRIs have global scope by design. Thus, two different appearances
>> >>     of an IRI denote the same resource. Violations of this principle
> may
>> >>     lead to interoperability problems or inconsistencies when, e.g.,
>> >>     using data from multiple sources.
>> >>
>> >> Would that address your concerns?
>> >
>> > That comes *very* close to addressing my concerns.  A slight tweak to
>> > the bullet item would do it:
>> >
>> >   . IRIs have global scope by design. Thus, two different appearances
>> >     of an IRI are intended to denote the same resource. Violations
>> >     of this principle may lead to interoperability problems or
>> >     inconsistencies when, e.g., using data from multiple sources.
>>
>> >> IRIs have global scope by design. Thus, two different appearances
>> >>     of an IRI should denote the same resource.  <etc.>
>
> I find this similarly confusing than David's proposal. If IRIs have global
> scope by design then all other cases are violations. "Should" or "are
> intended to" is, IMO, too weak in this case. Just because that constraint is
> sometimes violated in practice doesn't mean that the spec should be written
> in such a way. All RDF 1.1 specifications are based on this fundamental
> property of IRIs. If we make this too weak here, all other specifications
> are affected by this as well. The <> IRI in the following two triples MUST
> be the same in order for the two triples to make any sense:
>
>   <> rdf:type foaf:Person .
>   <> foaf:name "Markus".
>

The RDF-1.1 document format specifications should not require the
world to be free of all errors to be internally consistent. If they
appear in different RDF-1.1 Datasets, RDF triples should be free to
associate separately without causing a logical "MUST" failure due to
their simple existence. The statements above are both in the default
RDF-1.1 Dataset, so they would be taken as closely linked in the
absence of any other contextual information such as a user assigning
RDF-1.1 Datasets to them locally for any purpose (such as statement
provenance or Dataset provenance).

Writing the core specification in a strict manner should also not be
necessary to allow document formats that have slightly tighter
intended audiences (for example for those of the Linked Data
persuasion) from highly recommending that people do not intentionally
reuse URIs to refer to different real world entities. If someone wants
to accept statements into their system which reuse URIs to mean
different things (particularly in the case of different RDF-1.1
Datasets) that should be up to them to manage the statements to make
their system and their exposed statements consistent from the relevant
users points of view. Forcing them to be in violation of the base
RDF-1.1 Concepts model doesn't allow for explorations into the area of
how to manage URI meaning change with reference to RDF, for instance.

A reference to RDF-1.1 Datasets that had a SHOULD requirement with
respect to URI correference within a Dataset, and another statement
that used "intended" (or other non-RFC2119 wording) for global scope
works for me. There is no reason or rationale (from my experience) to
be locally managing a single RDF-1.1 Dataset and have the same URI
mean different things in neighbour triples.

I have already been using RDF-1.1 Datasets (previously based on SPARQL
GRAPH) versioning to concurrently manage OWL Ontologies within a
single system where the meaning of a URI could legitimately be
concretely changed between graphs, particularly versions of the
ontologies that are separated by numerous intervening minor changes
that together add up to a significant change. Users simply need to
understand the system design (based on OWL VersionIRI and RDF Quads
file formats) to avoid joining particular RDF-1.1 Datasets into a
single RDF Graph (typically based on the OWL OntologyIRI. Not being
completely sure about URI coreference between RDF-1.1 Datasets is not
the only reason to avoid naively joining arbitrary RDF-1.1 Datasets
together into a single graph. Updates to an RDFS:label (for instance)
to correct a spelling mistake would pollute the resulting graph in a
naive join.

Thanks,

Peter

Received on Monday, 16 December 2013 05:31:04 UTC