Hi Ivan,
> Some comments and clarification to this discussion from the
> Semantic Web community perspective.
Thanks very much for the comments, and for engaging with the RDFa folk
on this topic.
> 1. There is a set of strong voices in the current development
> around HTML (commonly referred to as HTML5) that try to
> dismiss the usage of profiles for HTML. Personally I think it
> is a mistake, but the major argument is that, until now, that
> attribute has not been used. I think your usage of the
> profile attribute _is_ the right choice, but I thought I
> should bring this to your attention.
OK, thanks. Yes, I was vaguely aware of the discussions around the
profile attribute, though I hadn't been tracking them in detail.
Speaking only for myself, I tend to agree that the profile attribute is
useful/necessary - and I don't see how things like "microformats" can
work in a global context without the equivalent of a "profile" to
"anchor" them.
(One alternative suggestion I've seen is to introduce a convention
something like
<link rel="profile" href="" />
with "profile" becoming one of the "built-in" link types defined by
HTML?)
> To be more proactive: I think the DCMI community should
> clearly raise its voice on the appropriate fora requiring
> that the profile attribute should _not_ be removed. This is a
> large, well, potentially huge:-) community, and its voice
> should be heard.
>
> (I will certainly forward your mail to the W3C staff contact
> for information. But the voice of a community is to be taken
> much more seriously).
OK, I'm a bit unsure of the procedures, but I guess if some sort of
collective representation is to be made, we could develop a form of
words here, and then Mikael & Tom as chairs could submit it on behalf of
the DC Architecture Forum? (Or if more, ahem, clout was desirable, ask
the Directorate to to do so?)
> 2. I have some comments and clarification on the RDFa issues
> below, but I mixed it with your questions. I hope those are helpful.
>
> Pete Johnston wrote:
>
> >
> > DC-HTML & RDFa
> > ==============
> >
> > What this new draft _doesn't_ address is any RDFa [4]
> interpretation
> > of an XHTML 1.0/1.1 doc using this profile.
> >
> > I must admit I'm still a bit unclear about how RDFa applies to XHTML
> > 1.0/1.1 docs. But my understanding (and I could be wrong
> about this!)
> > is that RDFa will not be defined as an X/HTML metadata profile, so
> > there will not be a profile URI for RDFa.
>
> This is not yet decided, and the RDFa group is still discussing it.
> There are serious arguments to define such profile.
Right.
> > However - at least in
> XHTML 1.1 -
> > there will be some other "hook"/"trigger" to signal that an
> XHTML 1.1
> > doc contains RDFa - a reference to a specific DTD in the DocType
> > declaration, I think?
> >
>
> Yes, a specific DTD is in development right now. But, as I
> said, this issue is not yet decided.
OK.
> > If I'm wrong about that, and if an RDFa processor _is_ going to
> > extract triples from an XHTML 1.0/1.1 doc regardless, then,
> given that
> > RDFa uses a QName-like convention based on XML Namespaces for
> > representing URIs, and this profile (and eRDF) uses a different
> > convention, I'd expect an RDFa processor to generate some rather
> > nonsensical triples e.g. given
> >
> > <meta name="dc.title" content="My title" /> <link rel="dc.creator"
> > href="http://example.org/Fred" /> <link rel="schema.dc"
> > href="http://purl.org/dc/elements/1.1/" />
> >
> > a GRDDL processor using the dc-html profile transform would generate
> >
> > <> dc:title "My title" .
> > <> dc:creator <http://example.org/Fred> .
> >
> > but an RDFa processor would generate (I think?)
> >
> > <> <http://www.w3.org/1999/xhtmldc.creator>
> <http://example.org/Fred> .
> > <> <http://www.w3.org/1999/xhtmlschema.dc>
> > <http://purl.org/dc/elements/1.1/> .
> >
>
> In the present situation that is indeed correct:-( I will
> forward your questions/comments to the RDFa group on this.
>
> > (I think the meta element would be ignored because RDFa uses a
> > different attribute for the predicate URI.)
> >
>
> I presume you refer to the 'name' attribute in the <meta>
> element.
Yes, that's what I was referring to. I was assuming an RDFa processor
would not generate a triple from a meta element in the absence of the
property attribute.
> Yes, at the moment, the 'property' attribute is used
> (with 'content'). Again, I will forward this issue to the
> RDFa community.
>
> > But I'm hoping that my concern here is without foundation,
> and an RDFa
> > processor _does_ need some hook before it goes to work on an XHTML
> > 1.0/1.1 doc, and so it will _not_ generate those spurious triples.
> >
>
> Hm. I am not sure the two are so clearly related. On the one
> hand, one of the issue is whether an HTML file is processed
> via an RDFa processor or not. But even if yes, there are some
> syntactic differences. The 'a.b'
> notation for a dublin core term is indeed not recognized by
> the current RDFa spec, regardless of the other issue.
Yes, understood.
> > But I suppose this begs the larger question of whether DCMI should
> > recommend shifting from this current approach (an X/HTML metadata
> > profile compatible with the eRDF profile and accessible to a GRDDL
> > processor) to an explicitly RDFa-based approach.
> >
> > Given that RDFa is still under development at this point in
> time, I'm
> > hesitant to recommend that change right now, and I think there is
> > considerable value in a GRDDL-able profile.
> >
>
> Just to make it clear how an RDFa-type profile would look
> like (I randomly picked example 9 from your document)
[snip]
> Ie, the difference is using ':' in the <link> and <meta>
> elements, using the @property instead of the @name, and
> using a namespace declaration instead of the <link
> rel="schema.DCTERMS".
Yes, agreed.
> I wonder whether this would be a major shift for DCMI; I
> cannot judge that. Note that it is absolutely no problem
> adapting a GRDDL script to follow this RDFa syntax, so I am
> not sure that argument is justified.
Yes, I understand that a transform can just as easily do the
QName-like-name-to-URI-via-XML-Namespace mapping as it can do the
dotted-name-to-URI-via-schema.XX mapping, and either/both could be used
as the basis of a GRDDL profile transformation.
I think my main concern was the profile/"trigger" question for RDFa: how
to signal to an RDFa processor that in document A, my rel attribute
values were QName-like-names constructed with the intent that they were
to be mapped to URIs via concatenation with XML Namespace names, but in
document B, they are plain strings and there is no intent that they
should be mapped to URIs via concatenation with the default XML
Namespace.
> > But at some point in the future once RDFa is done, it may be worth
> > producing a separate note on encoding DC metadata using RDFa.
> >
>
> Another possibility is to propose a syntax that is upright
> compatible with RDFa; this can be easily used and adapted
> with GRDDL...
OK, I think revising the syntactic conventions is an option we could
consider.
To make the DCMI convention compatible with RDFa, I think there are
three aspects where change would be required:
(i) the use of a colon (":") rather than the period (".") as the
separator in what I called in the draft "DC-HTML Prefixed Names";
(ii) the use of XML Namespace declarations, rather than the
link[@rel='schema.XX'] convention, as the basis for mapping the (now
colon-separated) "DC-HTML Prefixed Names" to URIs
(iii) the use of the meta/@property attribute rather than meta/@name for
the predicate
I suspect (i) wouldn't be too much of a problem (though that's just a
personal opinion!) i.e. I think people would be reasonably happy to
adopt the colon (":") rather than the period (".") as the separator in
the proposed profile. (Though N.B. that would still leave a lot of data
out there using the existing convention, so it seems to me two distinct
profiles would be required to distinguish the two cases.)
On (ii), one issue I see here is that DCMI's
dotted-name-to-URI-via-schema.XX convention was designed to be usable in
HTML as well as XHTML - I suspect it was designed before XML Namespaces
appeared on the scene (and quite possibly before XML). Sure, GRDDL etc
is XML-specific, and an XSLT transform wouldn't be applicable, but a
processor could still apply an algorithm to extract a DC metadata
description. So a convention based on XML Namespaces would be usable
only in XHTML, and not in HTML. I'm not sure how much of a problem that
would be, but I suspect there is still a significant constituency using
HTML, and an XML-specific solution might be problematic.
Similarly with (iii), the property attribute wouldn't be available in
HTML, and even in XHTML, if validity was required, then the use of the
property attribute would be valid only in XHTML 2 or in XHTML 1.1 with
the appropriate DTD.
> I hope this helps. I will also forward your mail to some
> other people at W3C to help moving things forward.
Thanks. I was away last week, but I've tried to catch up with the
discussion on [log in to unmask]
http://lists.w3.org/Archives/Public/public-rdf-in-xhtml-tf/
I think I'd agree with the view that trying to extend RDFa either to
incorporate the dotted syntax and the schema.XX prefixed name->URI
convention would be the wrong way to go, and I hadn't intended to
suggest that as an option.
However, I think Ben Adida's message here
http://lists.w3.org/Archives/Public/public-rdf-in-xhtml-tf/2007Aug/0119.
html
suggests a solution - augment the profile with an hGRDDL transform so
that for the XHTML case we can map to an XHTML document which is
RDFa-friendly - which (I think?) gives us "the best of both worlds" in
that
(a) no special casing is required in RDFa, and
(b) DCMI can define a set of conventions that can be deployed both in
HTML and XHTML (because they don't depend on XML Namespaces)
I think this will depend on ensuring that an RDFa processor will always
apply an hGRDDL transform before applying the native RDFa extraction
algorithms (i.e. to avoid the generation of the spurious triples with
predicates like http://www.w3.org/1999/xhtmldc.creator and
http://www.w3.org/1999/xhtmlschema.dc as above), but as long as that is
the case, then I _think_ we can have a profile which uses the
conventions proposed and which is also "RDFa-friendly" (via hGRDDL).
(I'm a bit unclear about the status of hGRDDL at this point in time, I
must admit.)
We may still want to consider changing the separator in what I called
"DC-HTML Prefixed Names" from a period to a colon? I'm not sure whether
that would be a good thing to do or not: given that they aren't QNames
and the prefixes won't be mapped to XML Namespace Names, it may be
confusing to make that change.
Cheers
Pete
---
Pete Johnston
Technical Researcher, Eduserv Foundation
Web: http://www.eduserv.org.uk/foundation/people/petejohnston/
Weblog: http://efoundations.typepad.com/efoundations/
Email: [log in to unmask]
Tel: +44 (0)1225 474323
|