Re: [SKOS] "SKOS RDF schema"

Hello Alistair,

>From one random W3C spec found by Google [1]:

normative

    required for conformance

    Note 1: One may conform in a variety of well-defined ways to this document.

    Note 2: Content identified as "informative" or "non-normative" is never required for conformance.


informative

    for information purposes and not required for conformance

    Note: Content required for conformance is referred to as "normative."



You can argue that this is not 100% clear in our case, as if we require conformance with OWL-Full ontology, we in fact also require conformance with the OWL-DL one (as it is a subset of it).
But from a document writing (and reading!) perspective it may matter: the only formal conformance condition we define (in section 1.8) is the one wrt. the OWL-Full ontology, which as as you say is the maximal set of axioms that captures the complete data model, given current (OWL) ontology modelling limitations.

One first argument to make the OWL-DL subset informative is that it is a subset, and we don't want people to believe that it is "as good" as the other model.
Another one is that even if we decided to highlight this OWL-DL subset as more interesting one, it would be cumbersome making it a "norm". there are several options to design the OWL-DL ontology from the OWL-Full one, and I think we don't have any scientific or practical ground to say that one should be preferred over another.

Now, I believe that the normative/informative distinction per se is enough for the reader to distinguish what we want to formally state as part of the norm from the rest. But I could live very well with the Reference including some more motivation (in the spirit of what we've written in this thread, in these last two mails, if the WG feel that they're ok): I don't think it makes that many words ;-)


[1] http://www.w3.org/TR/WCAG20/


> hi antoine,
> 
> On Wed, Apr 15, 2009 at 05:36:07PM +0200, Antoine Isaac wrote:
>> Hi Alistair,
>>
>> On the "normative" aspect. Isn't the typical W3C distinction between "normative" and "informative"?
>>
>> We could then have:
>> - a normative complete spec of thedata  model, defined by the vocabulary + different SXX axioms
>> - a normative machine-readable spec for RDF applications, which corresponds to a subset of the data model: the OWL-Full ontology
>> - an informative machine-readable spec for these RDF applications that need OWL-DL compliance.
>>
>> We could introduce the last spec as an informative annex. And there we could warn that the OWL-DL specification provided, corresponding to a subset of the subset defined by the OWL-Full ontology, will allow to detect fewer cases of inconsistency with the SKOS model, as defined in 1.8 (which could then be left unchanged).
>>
>> Would that make sense?
> 
> yes, intuitively that sounds good.
> 
> but i'd still like to clarify what we really mean by "normative" and
> "informative" here, and what it is that we're trying to achieve by
> applying these (or other) labels to the owl full and owl dl
> schemas. why isn't the owl dl schema also normative?
> 
> if you simply want to convey the message that the owl full schema
> should be preferred if possible, because it supports more entailments
> (including inconsistency entailments) than the owl dl schema, then i
> think you have to say that in so many words -- i don't think that
> message is conveyed by a normative/informative distinction.
> 
> cheers,
> 
> alistair
> 
>> Antoine
>>
>>> hi antoine
>>>
>>> On Tue, Apr 14, 2009 at 02:52:20PM +0200, Antoine Isaac wrote:
>>>> Hello,
>>>>
>>>> I would support the general scheme, especially considering the distinction SKOS/SKOS-XL: I also think it would be confusing to mix both, while they don't have the same status!
>>>>
>>>> Comments on some more precise aspects:
>>>>
>>>> - we should not forget about Tom's other recommendations on the content of documents (explaining the negotiation recipe used, using natural language labels of voc elements in HTML docs). These are very useful addition to the existing docs!
>>> yes, definitely.
>>>
>>>> - wrt. DL versions: +1 for refering to them from the documentation. But I would like them to be marked as "optional" or something that makes it clear that the "true" SKOS spec is the complete one, the DL one being only offered by the WG as a resource convenient for specific applications. A bit like SKOS-XL...
>>> tom points out that both the current rdf schema, and an owl dl prune
>>> of that schema, are sub-sets of the complete skos data model as
>>> defined in the skos reference.
>>>
>>> the owl wg [3] asked us to clarify that the current rdf schema is a
>>> "normative subset" of the skos data model, and that's what the skos
>>> reference currently says.
>>>
>>> now i'm trying to figure out 
>>>
>>>  * what "normative subset" means,
>>>
>>>  * whether it makes sense and/or is helpful to describe both the skos
>>>    rdf schema and an owl dl prune as "normative subsets" of the skos
>>>    data model, and if so,
>>>
>>>  * how to indicate that the owl full rdf schema is in some way more
>>>    "important" or less "optional" than the owl dl prune, or
>>>
>>>  * what "non-normative subset" might mean
>>>
>>> [1] glosses "normative" as: Text in a specification which is
>>> prescriptive or contains conformance requirements.
>>>
>>> one difficulty here is that section 1.8 of the skos reference says:
>>> "This specification does not define a formal notion of
>>> conformance. However, an RDF graph will be inconsistent with the SKOS
>>> data model if that graph and the SKOS data model (as defined formally
>>> below) taken together lead to a logical contradiction." [4]
>>>
>>> wordnet [2] defines "prescriptive" as: (pertaining to giving
>>> directives or rules) "prescriptive grammar is concerned with norms of
>>> or rules for correct usage"
>>>
>>> so you could say that...
>>>
>>> the skos reference contains a prescription of the skos data model. the
>>> main (owl full) skos rdf schema is also a prescription of the skos
>>> data model, albeit an incomplete prescription. an owl dl prune of the
>>> skos rdf schema is also a prescription of the skos data model, again
>>> incomplete, and even less complete than the owl full schema. but all
>>> three documents are prescriptive, and hence by [1] are normative.
>>>
>>> but then by that reasoning, anything which states some subset of the
>>> skos data model, either as prose or as rdf triples, is normative.
>>>
>>> i.e. to take it to an extreme, the triple
>>>
>>> skos:Concept rdf:type owl:Class .
>>>
>>> states a subset of the skos data model, and hence is normative.
>>>
>>> and i still don't have a clear notion of what "normative subset"
>>> means.
>>>
>>> does it makes sense to say the following...
>>>
>>>  * the skos reference contains a complete normative specification of
>>>    the skos data model
>>>
>>>  * the skos rdf schema (owl full) is a normative specification of
>>>    _the_ maximal subset of the skos data model that can be expressed
>>>    as rdf triples using rdf, rdfs and owl vocabularies.
>>>   * the skos rdf schema (owl dl prune) is a normative specification of
>>>    _a_ subset of the skos data model than can be expressed as rdf
>>>    triples using rdf, rdfs and owl vocabularies, and that is also a
>>>    valid owl dl document.
>>>
>>> ...?
>>>
>>> perhaps it would help to modify skos reference section 1.8 to rather
>>> say something like:
>>>
>>> "An RDF graph is not consistent with the SKOS data model if that graph
>>> and the SKOS data model (as defined formally below) taken together
>>> lead to a logical contradiction. Any RDF graph that is not consistent
>>> with the skos data model does not conform to this specification."
>>>
>>> ...? 
>>>
>>> but maybe that doesn't really change anything?
>>>
>>> i'm not sure what to do here.
>>>
>>>> - about the http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos "generic document", I am quite puzzled. We say nothing about it in the spec, it is not needed for content negociation as far as I understand, and yet it exists :-) If it is a mere accident, is there a way to hide it well under the carpet, or even turning off the multiviews feature that leads to such a behavior?
>>> fwiw i could live with either turning off multiviews or having a
>>> generic document by accident.
>>>
>>> cheers,
>>>
>>> alistair
>>>
>>> [1] http://www.w3.org/QA/glossary
>>> [2] http://wordnetweb.princeton.edu/perl/webwn?s=prescriptive
>>> [3] http://www.w3.org/2006/07/SWD/track/issues/154
>>> [4] http://www.w3.org/TR/2009/CR-skos-reference-20090317/#L434
>>>
>>>
>>>
>>>> Cheers,
>>>>
>>>> Antoine
>>>>
>>>>
>>>>> hi tom,
>>>>>
>>>>> i'd be happy with the following setup for PR, based on sticking close
>>>>> to established practice e.g. OWL and on making the least change to our
>>>>> own practices so far, although i'm happy to consider alternatives if
>>>>> anyone has a better idea...
>>>>>
>>>>>
>>>>> == SKOS ==
>>>>>
>>>>> we publish a "SKOS Namespace Document HTML Variant", at a url like:
>>>>>
>>>>> [a] http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos.html
>>>>>
>>>>> we publish a "SKOS Namespace Document RDF/XML Variant", at a url like:
>>>>>
>>>>> [b] http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos.rdf
>>>>>
>>>>> we publish a "SKOS RDF Schema (OWL 1 DL Sub-set)" at a url like:
>>>>>
>>>>> [c] http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos-owl1-dl.rdf
>>>>>  we make sure [a] has links directly to [b] and [c].
>>>>>
>>>>> in an appendix to the SKOS Reference, we state that the document at
>>>>> [b] should be considered to be a part of the SKOS Reference, and hence
>>>>> a part of the Recommendation. the appendix also states that [b] is a
>>>>> normative subset of the SKOS data model.
>>>>>
>>>>> in a separate appendix, we do the same for [c] (i.e. state that it is
>>>>> a part of the Recommendation, and that it is a normative subset of the
>>>>> SKOS data model).
>>>>>
>>>>> we *do not* include the rdf/xml content of [b] or [c] explicitly in
>>>>> these appendices, unless absolutely necessary to confer recommendation
>>>>> status to the rdf/xml content. (it seems a bit crazy to do what owl 1
>>>>> did and include the content of [b] or [c] explicitly in appendices to
>>>>> the skos reference, and hence duplicate the rdf/xml content in two
>>>>> places ... although if this is the only way we could resolve the
>>>>> status issue, i could live with it.)
>>>>>
>>>>> we do recipe 3 redirects from <http://www.w3.org/2004/02/skos/core>
>>>>> ("SKOS Vocabulary Namespace URI") to [a] and [b].
>>>>>
>>>>> we redirect from <http://www.w3.org/2004/02/skos/core.rdf> ("Latest
>>>>> SKOS Namespace Document RDF/XML Variant") to [b]. this link is just a
>>>>> convenience shortcut for developers, we mention it on the SKOS web
>>>>> site but we otherwise don't include it in formal documentation.
>>>>>
>>>>> because of the way that [a] and [b] are published on an apache server
>>>>> with multiviews enabled, there is also another url:
>>>>>
>>>>> <http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos>
>>>>>
>>>>> which is effectively a "generic document", but this is accidental and
>>>>> not by design. If we need to refer to this in discussion, we could
>>>>> call it the "SKOS Generic Namespace Document", but it is probably
>>>>> better not to mention this link anywhere either in formal
>>>>> documentation or on the web site, as it may cause confusion.
>>>>>
>>>>> alternatively, we disable multiviews for the entire directory, and so
>>>>> the URI above will 404. this might be safer, and would certainly avoid
>>>>> confusion over what cool URI pattern we're doing.
>>>>>
>>>>> a second alternative would be to switch from doing recipe 3 redirects
>>>>> to doing 303 redirects to one generic document (see
>>>>> <http://www.w3.org/TR/cooluris/#r303gendocument>), but it's a bit late
>>>>> in the game to switch our practices like this, and would require some
>>>>> testing and a rethink of inter-document links.
>>>>>
>>>>>
>>>>> == SKOS XL ==
>>>>>
>>>>> we publish a "SKOS XL Namespace Document HTML Variant", at a url like:
>>>>>
>>>>> [d] http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos-xl.html
>>>>>
>>>>> we publish a "SKOS XL Namespace Document RDF/XML Variant", at a url like:
>>>>>
>>>>> [e] http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos-xl.rdf
>>>>>
>>>>> we publish a "SKOS XL RDF Schema (OWL 1 DL Sub-set)" at a url like:
>>>>>
>>>>> [f] http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos-xl-owl1-dl.rdf
>>>>>
>>>>> which imports [c] instead of <http://www.w3.org/2004/02/skos/core> but
>>>>> is otherwise the same as [e].
>>>>>
>>>>> we make sure [d] links directly to [e] and [f].
>>>>>
>>>>> in an appendix to the SKOS Reference, we state that the document at
>>>>> [e] should be considered to be a part of the SKOS Reference, and hence
>>>>> a part of the Recommendation. in a separate appendix, we do the same
>>>>> for [f]. we *do not* include the rdf/xml content of [e] or [f]
>>>>> explicitly in these appendices.
>>>>>
>>>>> we do recipe 3 redirects from <http://www.w3.org/2008/05/skos-xl>
>>>>> ("SKOS XL Vocabulary Namespace URI") to [d] and [e].
>>>>>
>>>>> we redirect from <http://www.w3.org/2008/05/skos-xl.rdf> ("Latest
>>>>> SKOS XL Namespace Document RDF/XML Variant") to [b]. this link is just a
>>>>> convenience for developers, we mention it on the SKOS web site but we
>>>>> otherwise don't include it in formal documentation.
>>>>>
>>>>> because of the way that [d] and [e] are published on an apache server
>>>>> with multiviews enabled, there is also another url:
>>>>>
>>>>> <http://www.w3.org/TR/2009/PR-skos-reference-2009XXXX/skos-xl>
>>>>>
>>>>> which is effectively a "generic document", but this is accidental and
>>>>> not by design. If we need to refer to this in discussion, we could
>>>>> call it the "SKOS XL Generic Namespace Document", but it is probably
>>>>> better not to mention this link anywhere either in formal
>>>>> documentation or on the web site, as it may cause confusion.
>>>>>
>>>>> alternatively, we disable multiviews for the entire directory, and so
>>>>> the URI above will 404.
>>>>>
>>>>> ...
>>>>>
>>>>> how does that all sound?
>>>>>
>>>>> tom suggests combining SKOS and SKOS XL namespace documents into a
>>>>> single document, but i would rather keep them separate, mainly because
>>>>> having them separate reinforces the fact that XL is an optional
>>>>> extension to SKOS, and means that people coming first to SKOS don't
>>>>> have to deal initially with the extra layer of complexity.
>>>>>
>>>>> cheers,
>>>>>
>>>>> alistair
>>>>>
>>>>>
>>>>>
>>>>> On Mon, Apr 06, 2009 at 07:34:33PM +0200, Thomas Baker wrote:
>>>>>> Dear all,
>>>>>>
>>>>>> I'm looking at [1], also known as [2].
>>>>>>
>>>>>> [1] http://www.w3.org/TR/2009/CR-skos-reference-20090317/skos.html
>>>>>> [2] http://www.w3.org/TR/2009/CR-skos-reference-20090317/skos
>>>>>>
>>>>>> This document calls itself "SKOS Simple Knowledge
>>>>>> Organization System RDF Schema". 
>>>>>>
>>>>>> One obvious problem is that the document is clearly an HTML
>>>>>> page and not an RDF schema. It may seem obvious to us that it
>>>>>> is really _about_ the RDF schema, but I can well imagine this
>>>>>> causing some confusion.
>>>>>>
>>>>>> In the section of that document labeled "SKOS RDF Schema", then,
>>>>>> there is a link labeled "SKOS RDF Schema" which points to
>>>>>>
>>>>>> [3] http://www.w3.org/2004/02/skos/core
>>>>>>
>>>>>> which however -- since I am reading it in a browser --
>>>>>> content-negotiates back to [1]! 
>>>>>>
>>>>>> However, the section does provide an additional link directly
>>>>>> to the schema itself ("download the RDF schema without content
>>>>>> negotiation") [4].
>>>>>>
>>>>>> [4] http://www.w3.org/2004/02/skos/core.rdf
>>>>>>
>>>>>> Given the title and introduction, then, the reader does not
>>>>>> necessarily expect to find the _contents_ of the RDF schema by
>>>>>> scrolling down one screen.
>>>>>>
>>>>>> I find this all delightfully confusing... :-)
>>>>>>
>>>>>> Some issues and suggestions for discussion:
>>>>>>
>>>>>> -- Maybe call the Web document [1-2] "Contents of the RDF    
>>>>>> schema for Simple Knowledge Organization System (SKOS)"    or 
>>>>>> "Expressing SKOS in RDF" - in effect, anything but
>>>>>>    "SKOS RDF schema" :-)
>>>>>>
>>>>>> -- There should be a link labeled "SKOS RDF schema", but it    
>>>>>> should link to [4] directly, not via content negotiation from
>>>>>>    [3].  The surrounding text, however, should describe the     
>>>>>> content negotiation mechanism clearly.
>>>>>>
>>>>>> -- Status of the document [1-2]: I'm not sure we have properly
>>>>>>    resolved this question. I think it would help the reader if
>>>>>>    this document describes itself as a readable ready-reference
>>>>>>    page which excerpts the contents of the RDF schema for SKOS,
>>>>>>    and that the RDF schema for SKOS is itself a formalisation
>>>>>>    of a subset of the semantic conditions described in SKOS
>>>>>>    Reference, and that the SKOS Reference has the status of W3C
>>>>>>    [Candidate] Recommendation.
>>>>>>
>>>>>>    Like the OWL ontology [5], the RDF schema for SKOS does not
>>>>>>    assert any status for itself, though in the case of OWL,
>>>>>>    the contents of the schema are replicated in an appendix to
>>>>>>    the Recommendation document [6], arguably conferring on the
>>>>>>    schema itself a status of Recommendation, assuming it merely
>>>>>>    replicates the text in the appendix.
>>>>>>
>>>>>>    [5] http://www.w3.org/2002/07/owl
>>>>>>    [6] http://www.w3.org/TR/owl-ref/#appB
>>>>>>
>>>>>>    This text provides an opportunity to clarify that the SKOS
>>>>>>    Reference is the [Candidate] Recommendation and that [1],
>>>>>>    [2], and [4] are all excerpts of that Recommendation.
>>>>>>    (The text should point out that it is not possible to
>>>>>>    express all of the statements of the SKOS data model as RDF
>>>>>>    triples and thus the schema forms a "normative subset" of the
>>>>>>    specification.)
>>>>>>
>>>>>> -- I'm wondering if this document [1-2] might be a good    place 
>>>>>> to introduce and link any DL versions of SKOS as per Sean's
>>>>>>    discussion in [7].  The document could briefly explain the 
>>>>>> need    to have a DL ontology in some contexts, describe the 
>>>>>> algorithm
>>>>>>    by which some of the axioms in [4] are "thrown away" (or filtered
>>>>>>    out).  In a way, the DL schema is arguably just another subset
>>>>>>    of the semantic conditions described SKOS Reference, albeit a  
>>>>>>   deliberately lossy one.  Grouping points to the RDF, HTML, and 
>>>>>> DL    excerpts together in one document would be helpful; and 
>>>>>> where else    to do that but in this document [1-2]?  (This 
>>>>>> assumes we want to    publish, or indeed say anything at all 
>>>>>> about, a DL version - also an
>>>>>>    issue for discussion.)
>>>>>>
>>>>>>    [7] http://lists.w3.org/Archives/Public/public-swd-wg/2009Mar/0071.html
>>>>>>
>>>>>> -- The HTML representation [1-2] should include labels.
>>>>>>
>>>>>> -- If scripts were used to extract [1-2] from SKOS Reference and 
>>>>>> to    generate the DL version, it might be useful to point to 
>>>>>> those scripts
>>>>>>    or include them in an appendix.
>>>>>>
>>>>>> -- Instead of replicating this construct separately for the XL  
>>>>>> namespace [8],    maybe SKOS and SKOS-XL could be usefully 
>>>>>> combined into one document
>>>>>>    "Expressions of SKOS Reference in RDF" (since SKOS Reference defines both
>>>>>>    the SKOS and XL namespaces).  Such a title would make clear 
>>>>>> that it    derives from SKOS Reference without implying that the 
>>>>>> document is itself
>>>>>>    an RDF schema.
>>>>>>
>>>>>>    [8] http://www.w3.org/2006/07/SWD/SKOS/reference/20081001/skos-xl.html
>>>>>>
>>>>>> If the document [1-2] were to do all of the above, might we
>>>>>> not want to formally approve it as a Working Group Note? That
>>>>>> would remove any confusion or ambiguity as to its status.  I   
>>>>>> hesitate because I'm not sure now under what circumstances a 
>>>>>> reader would normally click on or be redirected to this document
>>>>>> and whether, in those contexts, the reader might find it confusing
>>>>>> to encounter something with the status of Note.
>>>>>>
>>>>>> Tom
>>>>>>
>>>>>> -- 
>>>>>> Tom Baker <tbaker@tbaker.de>
>>>>>>
>>
>

Received on Thursday, 16 April 2009 08:51:50 UTC