Re: A question about referential opacity (again)

There is no restrictions on the sharing of resources between different 
interpretations in the RDF semantics.  Different interpretations of the same 
RDF graph can share resources.  Different interpretations of different RDF 
graphs in an RDF dataset can share resources.  Different interpretations of 
different RDF graphs that are not in an RDF dataset can share resources.

But there is nothing in the formal semantics of RDF graphs that depends on 
sharing or not sharing, except for the denotations of literals.  So no formal 
consequences would arise from forbidding interpretations sharing resources, 
again except for value spaces of datatypes.

On the other hand, much of the "intended" meaning of RDF graphs implies that 
different interpretations share resources, i.e., that the denotation of many 
IRIs are intended to be in some sense fixed between some interpretations.  One 
can even imagine a variation on RDF semantics where the denotation of an IRI 
is required to be the same in all interpretations.   This semantics would have 
to be somewhat unusual but it could be made to work.

Requiring denotation of IRIs to be fixed within an RDF dataset would likely 
also need an unusual semantics if it was to handle things like beliefs (or 
indeed any other kind of varying view of the identity of things in the world).

peter




On 10/23/23 09:00, Niklas Lindström wrote:
> On Sat, Oct 21, 2023 at 2:22 PM Peter F. Patel-Schneider
> <pfpschneider@gmail.com> wrote:
>>
>> It's important to be clear as to what is formal and what is informal in
>> discussions of this sort.
>>
>> Formally in the current semantics for RDF, all IRIs are mapped (via the map
>> IS) to resources by interpretations in RDF.  One generally says that an IRI E
>> denotes or refers to IS(E).
> 
> Is it correct that the interpretation I of a graph G in a dataset D
> must not by definition share its "domain/universe" (that is its
> non-empty set IR of resources) with another graph G' in D (or any
> other dataset)? That is, this is undefined (i.e. this is the lack of
> semantics for datasets)?
> 
> And if these universes may be different, is there possibly a different
> interpretation for each graph within the same dataset (or any
> combination of shared and isolated interpretations)?
> 
> And finally, would the definition of such shared or not shared
> interpretations constitute a semantics for datasets? (Possibly, but
> perhaps not necessarily, in conjunction with a definition for what the
> relation is between the name and the graph in a named graph pair?)
> 
> /Niklas
> 
>> A formal semantics that provides for referential opacity of IRIs generally
>> provides a different mapping (let's call it IS') for IRIs that occur in opaque
>> contexts, i.e., inside triple terms.   The details may differ, but the targets
>> of this mapping are usually either left unspecified or are to some synthetic
>> resources (such as copies of IRIs).
>>
>> So if one was to construct an interpretation in this sort of formal semantics
>> that actually included real cities in the world as resources and whose IS
>> mapping mapped IRIs that are generally accepted as names of cities to the
>> actual cities one would say that in transparent contexts, i.e., subjects,
>> objects, and properties of asserted triples, dbr:Linköping refers to the city
>> of Linköping but in opaque contexts, i.e., in triple terms, refers to
>> something else.  (It may be possible that some interpretations dbr:Linköping
>> in an opaque context does refer to the city, but the important point is that
>> there are interpretations where dbr:Linköping in an opaque context does not
>> refer to the city and that absent special constructs to force transparency on
>> opaque contexts there is no way to force an RDF graph to be false in all these
>> interpretations.)
>>
>> peter
>>
>> PS: The coordination group semantics uses a different mechanism entirely,
>> instead syntactically transforming graphs that contain triple terms to regular
>> RDF graphs.  Among other things, the opacity mechanism in this treatment
>> transforms IRIs in triple terms to literal strings.  So part of this semantics
>> is a relationship (similar to but not exactly denotes) from IRIs in opaque
>> contexts to sequences of Unicode code points.
>>
>> PPS:  There are other ways of obtaining opacity.  If the working group
>> switches to graph terms the kind of semantics described above might not be
>> adequate and some other treatment might have to be used.
>>
>>
>>
>> On 10/21/23 07:00, Thomas Lörtsch wrote:
>>> Hi,
>>>
>>>
>>> Enrico was kind enough to guide me through the work of the Semantics TF in a one-on-one TelCo a week ago. However, when I now look at my notes, I’m again confused.
>>>
>>> If I understood Enrico correctly then a referentially opaque IRI doesn’t refer to anything. However, it was my understanding of the CG report semantics that IRIs in quoted triples are interpreted, but strictly following the syntactic form. My reading of the unstar-mapping supports that intuition [1].
>>> To give an example, I understood referential opacity as meaning that "dbr:Linköping" and "DBR:LINKÖPING" both refer to the city of Linköping, and yet are not equal (and can not infered to be equal) because their lexical representation differs.
>>> But according to how I understood Enrico they don’t refer to anything.
>>>
>>> Was I wrong all along? Am I just not getting it and does there exist a world in which both interpretations are true? Or has the TF diverged from the CG? Or is there no consensus in the TF?
>>>
>>>
>>> Best,
>>> Thomas
>>>
>>>
>>>
>>>
>>> [0] https://github.com/w3c/rdf-star-wg/wiki/Semantics%3A-Behaviour-catalogue
>>> [1] https://w3c.github.io/rdf-star/cg-spec/2021-12-17.html#mapping
>>

Received on Monday, 23 October 2023 14:47:19 UTC