Re: Datatype maps

Thank you, Richard.

Regards,
Dave
--
http://about.me/david_wood



On Dec 11, 2013, at 13:57, Richard Cyganiak <richard@cyganiak.de> wrote:

> On 10 Dec 2013, at 13:13, Sandro Hawke <sandro@w3.org> wrote:
>> [taking this back inside the WG]
> 
> Here’s my take on Michael’s and Antoine’s threat of formal objection. I’d like to address a number of points:
> 
> 1. Does the abolition of datatype maps from Concepts simplify the specification set?
> 2. Does this change make RDF 1.1 incompatible with other specs, future or past?
> 3. Is the change unnecessary?
> 4. Which constituencies are affected by the change?
> 
> TL;DR: Antoine and Michael prioritise the interests of a small minority — writers of Semantics-related specifications — over the interests of all other constituencies. The editors of Concepts and Semantics should stick to their guns.
> 
> 
> 1 SIMPLIFICATION
> ===============
> 
> One of my editorial goals for RDF 1.1 Concepts was to move the list of supported XSD types out of Semantics and into Concepts. This, I think, is a major improvement in usability between the 2004 specification set and the 1.1 specification set.
> 
> In early drafts, we simply copied the formalism of a Datatype Map from Semantics into Concepts. This was a half measure. Datatype maps may or may not be useful in defining the formalism of Semantics, but they really have no place in Concepts. IRIs identify things. This is a ubiquitous concept in RDF, and it's entirely sufficient to explain the relationship of datatypes to their IRIs. I think we now do the Right Thing by simply stating that datatype IRIs denote datatypes, and then adding:
> 
> [[
> The list of datatypes supported by an implementation is determined by its recognized datatype IRIs.
> ]]
> http://www.w3.org/TR/rdf11-concepts/#section-Datatypes
> 
> This is a significant improvement over the state we had in 2004. It is also a significant simplification over the earlier drafts that talked about Datatype Maps, and that Antoine proposes to revert to.
> 
> Another goal was to improve the WG’s practices with regards to formally defined conformance criteria. The RDF 2004 specs are very poor in this regard, as is (sadly) most of the Semantic Web Activity’s output. RDF 1.1 is in much better shape. The change towards “recognised datatype IRIs” can further improve things going forward. Specifications which define new datatypes now can state in their Conformance section: “To conform to this specification, an implementation MUST recognise the IRI <xxx> as denoting the so-and-so datatype.” Previously, such specifications would have had to talk about constraints on datatype maps and entailment regimes to be formally correct.
> 
> Therefore, I find the argument that the latest draft simplifies the RDF 1.1 specifications absolutely justified.
> 
> My arguments above are about RDF Concepts, while Michael and Antoine are mostly concerned with the state of RDF Semantics. But as a matter of consistency between Concepts and Semantics, it makes sense for Semantics to define entailment in terms of the notion already present in Concepts (recognised datatype IRIs), rather than introducing a different notion (datatype maps).
> 
> 
> 2 COMPATIBILITY
> ==============
> 
> Does this change make RDF 1.1 incompatible? Let’s look first at existing specifications, and then at future specifications.
> 
> The argument that this change somehow makes the existing current versions of other specs incompatible with RDF 1.1 is, simply put, nonsense. These specs reference RDF 2004, not RDF 1.1. These are different languages, with quite a few differences, many of them larger than this one (that is, potentially user-visible, like the introduction of rdf:langString).
> 
> So let’s look at the argument that this change makes it harder to update some existing specifications in the future. I suppose this is about OWL's RDF-based semantics, SPARQL Entailment Regimes, and perhaps something in RIF. I will not get into arguments how likely updates to these specs are.
> 
> The argument seems very strange to me, given that future versions of these specs can easily re-introduce the concept of a datatype map as a bit of additional formalism, if that makes them easier. This could be a one-sentence change. Arguably, not even that one sentence is needed with the latest wording that Pat proposed.
> 
> Michael has suggested that it would be embarrassing to RDF-WG if other specifications found it necessary to re-introduced the formalism eliminated from RDF Semantics. I don’t know where he gets that idea. Moving a concept from one spec to another spec often improves factoring and is a net simplification.
> 
> 
> 3 IS THE CHANGE UNNECESSARY?
> =============================
> 
> No, as argued above.
> 
> But even if it was, that would be irrelevant. Arguments about necessity do not absolve the commentator from actually showing a negative impact on a constituency. If the change doesn’t cause harm, then it’s a matter of editorial discretion.
> 
> 
> 4 IMPACT ON CONSTITUENCIES
> ==========================
> 
> Michael initially threatened formal objection under the assumption that the change results in different entailments, which would indeed be a serious issue. I’m still not sure if he has accepted that he was wrong about this.
> 
> The change has no impact on entailments.
> 
> It has no impact on end users.
> 
> It has a positive impact on content authors and implementers (due to simplified specifications, as I argue above).
> 
> It has a positive impact on some specification writers (those who define new datatypes for use with RDF, as I argue above).
> 
> A constituency that arguably is negatively impacted are editors of hypothetical future specifications that reference RDF 1.1 Semantics, because they need to discuss and decide whether to remove datatype maps from their formalisms or not.
> 
> Surely it is correct for the RDF WG to prioritise the interests of content authors and implementers of its products over the interests of hypothetical editors of future specifications. I don’t think we should make RDF Concepts and RDF Semantics more complicated to save an editor of OWL3 a day of work.
> 
> Michael also argued that there is a negative impact on textbook authors, university lecturers, and perhaps their students. I’m not sure whether to believe this. Either way, unless they also fall into one of the groups above, their well-being is of no concern to RDF-WG or W3C.
> 
> 
> 5 SUMMARY
> ==========
> 
> I feel that Antoine and Michael, with their threat of formal objection, prioritise the interests of a tiny minority — writers of Semantics-related specifications — over the interests of all other stakeholders.
> 
> The HTML WG’s Priority of Constituencies is a good guideline here. I will slightly misquote it here:
> 
> [[
> In case of conflict, consider users over authors over implementors over specifiers.
> 
> In other words costs or difficulties to the user should be given more weight than costs to authors; which in turn should be given more weight than costs to implementors; which should be given more weight than costs to authors of the spec itself. Of course, it is preferred to make things better for multiple constituencies at once.
> ]]
> http://www.w3.org/TR/html-design-principles/#priority-of-constituencies
> 
> I therefore feel that the editors of Concepts and Semantics should stick to their guns, and the WG should proceed with the formal objections of Michael and Antoine on the record. I am certainly prepared to argue our case before the director.
> 
> 
> Best,
> Richard