On 11 Nov 2013, at 17:40, Christian Chiarcos <[log in to unmask]> wrote:
> ...(i) The general usecase would be a linking with an ontology, say, the
> SAWS ontology. Right now, it uses a TEI-specific formalism, which is
> fine (though possibly problematic as the construct may be in line with
> the TEI specifications, but not the RelaxNG Schema). But another
> approach would be to make use of RDFa, thereby allowing an RDFa
> processor to extract the RDF triples directly from the TEI -- without
> TEI-specific technology.
well, sort of. you can look for the RDFa things, but you still have to
understand the TEI context, surely, for it to to make any sense?
> (ii) The requirements for the host language are summarized under
> http://www.w3.org/TR/rdfa-core/#hostlangconf: the host language needs
> to provide a number of attributes, and these should not be namespaced.
> (iii) The attributes are under http://www.w3.org/TR/rdfa-core/#s_syntax
Adding those to the global attribute set available on all TEI elements
would be perfectly plausible, using a TEI customization. That could
be maintained alongside eg TEI + MathML.
> The benefit would be that RDF information from TEI could be directly
> processed with out-of-the-box tools. There are usecases for an LOD
> linking of TEI data, the SAWS usecase would be one, another one may be
> that the digital edition of, say, a printed lexicon or a linguistic
> corpus (in TEI, of course, to properly capture structure and/or
> layout), is directly augmented with semantic specifications relating
> it to one of the emerging data models for machine-readable lexicons or
> corpora in the (Linguistic) Linked Open Data world, say lemon
I can't help feeling that there would be redundancy. one would have a
TEI-specific set of semantic elements in their appropriate hierarchy,
from which one could derive LOD statements automatically; and then
one would add RDFa attributes which say the same thing explicitly. If
you start from a dumb,non-semantic, language like HTML then of course you
need the RDFa to add intelligence to the <div> and <span>s, but we have that
intelligence built into the element names and their attributes.
I am torn between three scenarios:
a) I use low-level TEI markup (<p>, <hi> etc) and add RDFa to provide the
information for LOD. but then why use TEI at all?
b) I write high-level native TEI, and derive RDF triples as a publishable output,
using the inherent knowledge in the TEI
c) I generate publishable HTML from my TEI, which exposes interesting things using
each of which has its upside and downside. what I think would be _wrong_
d) use high-level native TEI, and duplicate the semantics with added RDFa
as you say:
> The other disadvantage is that relational semantics can be expressed
> in different ways, then, in a pure TEI way, in a maxium-RDFa way or in
> a mixture of both.
I wonder if anyone can give a concrete example of how using RDFa
on top of TEI would work, _either_ because it would be much easier
to write, _or_ because it would be more expressive?
Director (Research) of Academic IT
University of Oxford IT Services
13 Banbury Road, Oxford OX2 6NN. Phone +44 1865 283431