Re: Ontology for Media Resource 1.0

Hi Felix,

On Sat, Jul 24, 2010 at 08:33:20AM +0200, Felix Sasaki wrote:
> This is because one use case for the mappings of existing
> formats to the "ma" vocabulary is to be information used by an API, as
> described in the "API for media resources" document, see
> http://www.w3.org/TR/2010/WD-mediaont-api-1.0-20100608/ . In that use case,
> you basically need to know about the mappings and apply them e.g. in API
> methods like this one
> http://www.w3.org/TR/2010/WD-mediaont-api-1.0-20100608/#contributor--interface.
>
> That is, you don't need any Semantic Web based machinery to implement this.
> Another use case is to provide a mapping using an RDF-based ontology. A task
> force within the working group is working on that, since there is a heavy
> demand for this use case as well. Nevertheless it is important that the
> working group produces a set of mapping which can be used for both use cases
> and which fulfills both needs - e.g. a browser-centered, let's say
> JavaScript-API and the application of the mappings for linked data
> scenarios. It is unfortunate that you don't see the RDF-based ontology yet,
> but it is on it's way.
>
> Finally let me emphasize that the key to the whole endavour is to get broad
> consensus about the mappings, no matter if they are expressed as a table or
> as RDF. So I encourage you to have a detailed look at the mappings and
> provide comments to the working group.

Thank you for bringing us up to date on this and putting the
mediaont endeavor into perspective.

I am interested to read in the Introduction [1] the rationale
for creating properties based on Dublin Core properties, only
more tightly constrained and under the control of the mediaont
editors [2].  

The more general issue here, as I see it, is what are the
tradeoffs for interoperability between simply "reusing"
existing external properties (by citing them) as opposed to
creating new properties related to the external properties
(the approach taken in mediaont).  

I suspect the answer is not black-and-white because, as the
Introduction points out, there are issues of constraints and
control when relying on external vocabularies, but on the other
hand, if every specification were to reinvent, say, "title",
resolving mappings of more "specialized" properties to more
"core-like" properties would involve a significant amount of
(human and machine) processing overhead.

As for reviewing the mappings: the mappings are currently
defined between elements of fixed "formats" -- e.g., for
Dublin Core, there are relevant mappings to be found in 4.2.2.3
(EBUCore), 4.2.2.8 (MediaRDF), and 4.2.2.16 (XMP) [2,3,4].

For the mappings to be effective in (and reviewable for use
in) linked data, the "properties" would need to be declared
specifically as RDF or OWL properties and the "mappings"
would need to be declared using triples. Section 4.2.1.3,
Mapping Expression [5], seems to say that mappings may
eventually be expressed using SKOS mapping relationships:

    The mapping expression corresponds to the concrete
    implementation or representation of the mappings defined
    in the previous paragraph, both at a semantic level
    and at syntactic one.  ... [SKOS] defines a vocabulary
    for representing Knowledge Organization Systems, such as
    vocabularies, and relationships amongst them. In SKOS the
    mapping properties that we take into account in the mapping
    table are expressed as: skos:exactMatch, skos:narrowMatch,
    skos:broadMatch and skos:relatedMatch. A future version
    of this specification MAY include additional information
    after interoperability and/or other feedback mechanisms
    have been completed.

I think we are touching here on another issue of general
importance for LLD XG -- best practice for expressing mapping
relationships in a linked data environment.  I'm wondering
if the mediaont group working on the RDF expression is
considering using more powerful equivalence assertions,
e.g. with rdfs:subPropertyOf or even owl:equivalentProperty?

On the other hand, if we were seeing here the start of
a trend towards use SKOS mapping properties for mapping
RDF properties, what assumptions are being made about how
consuming applications should process such triples when
merging or linking data?

In a word: What is the emerging best practice for expressing
mappings?  Is the existing arsenal of RDF, OWL, and SKOS
properties usable for mappings both rich enough (e.g.,
in differentiating between exact, close, more general,
more specific...) and semantically powerful enough (e.g.,
for inferencing) to meet requirements for mappings?

More immediately, unless I am missing something, the basis
on which members of this group might currently evaluate the
mediaont mappings from a linked-data perpective has not yet
been created (but is in the works?).

Tom

[1] http://www.w3.org/TR/mediaont-10/#introduction
[2] http://www.w3.org/TR/mediaont-10/#d0e5511
[3] http://www.w3.org/TR/mediaont-10/#d0e3064
[4] http://www.w3.org/TR/mediaont-10/#d0e9670
[5] http://www.w3.org/TR/mediaont-10/#mapping-expression

-- 
Thomas Baker <tbaker@tbaker.de>

Received on Saturday, 14 August 2010 15:29:21 UTC