Re: RDF Dataset semantics

On 09/17/2013 08:38 AM, Antoine Zimmermann wrote:
> Thanks for the comments.
>
> In order for me to address these, could you be more specific to what 
> sections you refer to?  Detailed answers inline.

Almost all these comments are about the first portion of the document.   I 
didn't see any section numbering when I viewed it.
>
>
> Le 17/09/2013 14:42, Peter F. Patel-Schneider a écrit :
>> This document promotes the notion that graph names denote. Technically
>> this is now benign, as in the new semantics all IRIs and blank nodes
>> denote, but there is the implicit assumption that graph names in RDF
>> datasets are supposed to denote something as graph names.  I'm not sure
>> just how to change the wording - "IRIs or blank nodes used as graph
>> names" is rather cumbersome.
>
> I'm not sure I understand well your concern. I have the impression that you 
> find the phrase "graph name" problematic. I find it too, but it's the way we 
> called this in concepts (the introduction of bnodes for graph names made it 
> much more difficult to write this document correctly).

I don't find the phrase "graph name" problematic at all.  What I do find 
problematic is the assumption that there needs to be some sort of account of 
the denotation of graph names.  The current account of RDF datasets doesn't 
need to talk about the denotation of graph names at all.
>
>>
>> There are lots of other natural choices for the denotation of graph
>> names (assuming that one wants these to denote something of interest).
>> They could denote the set of interpretations of the graph, they could
>> denote the mapping from the graph to its interpretations, they could
>> denote the set of graphs equivalent to the graph, they could denote the
>> inferential closure of the graph. Of course, these are all covered in
>> the second-last possibility in the document, but that is so general that
>> it could include something like the graph obtained by swapping subjects
>> and objects in the named graph, so it seems to me that some semantic
>> denotations should be called out to balance the numerous syntactic
>> denotations.
>
> Yes, there are many possible choices, I can't name them all. I especially 
> emphasise those that were discussed. I can rephrase to show that there are 
> many more options.

What I found problematic was the emphasis on a particular kind of denotation 
for graph names.   Including some semantics-based denotations would help.
>
>
>> The 2004 RDF semantics defines the meaning of sets of RDF graphs by
>> providing definitions of entailment from sets of RDF graphs to an RDF
>> graph.    The merge of a set of RDF graphs is then shown to have a
>> certain special relationship to this definition.  I can find no place in
>> the 2004 RDF semantics where merging is used to define the meaning of
>> RDF graphs.  The 2013 RDF semantics does not define entailment for sets
>> of RDF graphs because one has to first determine what to do with blank
>> nodes that are shared between the graphs.
>
> I see. I will rephrase the relevant paragraph.
>
>
>> It is misleading to imply that there has been a significant change in
>> the semantics of blank nodes between 2004 and 2013.
>
> I did not intend to convey this idea. I'll see how I can rephrase.
>
>
>>  In 2004 it was
>> assumed that blank nodes would not be shared between RDF graphs.
>
> I've never seen where this assumption is made in either RDF 1.0 Concepts or 
> semantics. In any case, it was very explicit that there are graphs that 
> share blank nodes.

The semantics was written in a way to allow for two graphs to share blank 
nodes, and concepts sort of followed along.  However, there was no way to get 
graphs to share blank nodes, and my view of the situation was that such 
sharing was considered exceptional.
>
>
>> However, the semantics needed to say something about what would happen
>> in this case, and the choice was to go with separating the
>> illegally-shared blank nodes.   (Perhaps a better choice would have been
>> to disallow this situation in the semantics.)
>>
>> There are conditions on the pairs in an RDF dataset that are not
>> captured in the document.
>
> You mean, that there cannot be two pairs with the same "graph name" in one 
> dataset? Ok, I'll fixed that.
>
OK
>
>
>> I would not say that it is problematic to provide a meaning for RDF
>> datasets.   There are very many unproblematic reasonable semantics for
>> RDF datasets.  It is just that many of these are incompatible so picking
>> one generally rules out others.
>
> Ok, bad formulation: it's problematic to provide a normative semantics, 
> precisely because it would make other unproblematic reasonable semantics 
> incompatible with the standard.

Much better.
>
>
>> It is not the case that the semantics for RDF datasets that ignores the
>> named graphs has "no added value".  The simple ability to carry around
>> the uninterpreted named graphs can be used to decorate RDF graphs.
>
> You're right. I noticed that there are several sentences I wrote in the 
> first version published (January 2013) that are overstated. I have to read 
> this through again and I welcome comments and suggestions for improvement.

Just don't disparage this particular solution.
>
>
>> I can't find a place where the actual meaning of RDF datasets is
>> mentioned or described.
>
> What do you mean? What is the actual meaning of RDF datasets?

Hmm.  I didn't include enough context here.   In any case, this comment is 
subsumed by the others in my message.
>
>
>
>> There should be a reference to the graph names paper by Jeremy et al.
>> where the paper is first mentioned.
>
> Oops, yes indeed.
>
>
>>
>> There are a number of minor grammatical issues that need to be fixed.
>
> As I said, I need to proof read this. If you have specific instances with 
> reference to the section, that'd help.
>
>
>
> Thanks a lot,
> AZ
>
>>
>> peter
>>
>>
>> On 09/17/2013 04:13 AM, Antoine Zimmermann wrote:
>>> I've updated dataset semantics [1].
>>>
>>> I especially invite Pat, Peter and certainly Sandro to take a look at
>>> it. I've tried to be comprehensive but there are many parameters that
>>> can vary and produce different formalisations, and different entailments.
>>>
>>>
>>> [1] https://dvcs.w3.org/hg/rdf/raw-file/default/rdf-dataset/index.html
>>
>>
>

Received on Tuesday, 17 September 2013 17:35:28 UTC