In yesterday's Provenance Task Group telecon we found ourselves talking about
DCAM [1]. One point of discussion was the analogy of DCAM to SKOS.
On January 5, Andy had written:
> So I think the pertinent question that needs to be answered pretty early on in
> the outer layers of Stuart's onion is "why should I invest time understanding
> the DCAM when I could be learning RDF/Linked Data/whatever instead?".
>
> If we compare the DCAM with, say, SKOS and ask the same kind of question the
> answer is more obvious I think - people need to understand both RDF and SKOS
> because SKOS gives them something useful in the area of 'vocabulary' handling
> that RDF on its own doesn't give them.
>
> The answer for the DCAM is much less clear except in terms of the original
> rationale for having the DCAM at all, i.e.
>
> "It provides an information model which is independent of any particular [DCMI]
> encoding syntax. Such an information model allows us to gain a better
> understanding of the kinds of [DCMI] descriptions that we are encoding and
> facilitates the development of better mappings and cross-syntax translations"
> ("[DCMI]" additions by me).
>
> which, unfortunately, is a very inward looking (and rather narrow) rationale
> that is unlikely (as history has shown us) to be of much widespread interest.
To which Kai had responded:
> [The] analogy to SKOS is perfect, because that was
> exactly how I started the RDF-based DCAM wiki page yesterday [1].
> Provide DCAM as a model for metadata just like SKOS is for vocabulary
> handling.
>
> [1] http://wiki.dublincore.org/index.php/DCAM_Revision_Tech
In yesterday's call, Kai elaborated on the notion of DCAM as an equivalent of
SKOS for metadata. I understood him to say that SKOS is an RDF vocabulary, but
one might also see it as an Abstract Model that could be used by people who do
not care about RDF.
This reminded me that in the Semantic Web Deployment WG, we did in effect try
to express a high-level "abstract model" for SKOS (a formulation I actually
helped write) [2]:
Using SKOS, _concepts_ can be identified using URIs, _labeled_ with lexical
strings in one or more natural languages, assigned _notations_ (lexical
codes), _documented_ with various types of note, _linked to other concepts_
and organized into informal hierarchies and association networks,
aggregated into _concept schemes_, grouped into labeled and/or ordered
_collections_, and _mapped_ to concepts in other schemes.
...summarizing the essence of SKOS in just one sentence. Arguably, this is the
sort of formulation -- one which does not itself even mention RDF but which
maps to RDF in the specification -- we could aspire to make for DCAM.
I cannot readily formulate one sentence that summarizes what I think DCAM can
offer, though it would perhaps be interesting to try. The story I have in mind
for DCAM might say that metadata uses items of information -- strings and URIs,
perhaps belonging to sets of strings or URIs (i.e., syntax or vocabulary
encoding schemes) -- to describe (make statements about) things of interest;
that it groups these items into Descriptions about one particular thing of
interest and groups related Descriptions into Description Sets, which are often
instantiated in implementations as "records".
How these items are used to make meaningful "statements" about things would be
the part that one inherits from RDF. DCAM, as I see it, can provide an
"interface" to underlying (meaningful) statements by specifying patterns of
information items grouped into Descriptions and Description Sets.
If that is what DCAM is, or should be, then I wonder whether we can specify
those patterns in enough detail to be useful as an interface to triples without
becoming too complicated. In 2007-2008, for example, it seemed reasonable to
translate "DCAM statements" about value resources using RDF statements with
rdf:value and literals or RDF statements with dcam:memberOf and vocabulary
encoding scheme URIs [3]. From the perspective of best practice, that looks
like an oversimplification. Today, one might want to consider using various
other properties in statements about a value resource -- rdfs:label,
skos:prefLabel, skos:notation, foaf:name, or dcterms:title... -- though perhaps
_not_ rdf:value [4]. Can a DCAM still be defined as an interface to triples as
straightforward as [4], or would it need to evolve in the direction of a more
complex and differentiated set of patterns?
For discussion on Monday's call (at 11:00 EST)...
Tom
[1] http://wiki.bib.uni-mannheim.de/dc-provenance/doku.php?id=minutes_2012_01_15
[2] http://www.w3.org/TR/skos-reference/
[3] http://dublincore.org/documents/dc-rdf/#sect-4
[4] http://www.w3.org/2011/rdf-wg/track/issues/27
--
Tom Baker <[log in to unmask]>
|