W3C home > Mailing lists > Public > public-media-annotation@w3.org > March 2011

Re: mapping, RDF and test-cases

From: Pierre-Antoine Champin <pierre-antoine.champin@liris.cnrs.fr>
Date: Wed, 16 Mar 2011 10:40:05 +0100
Message-ID: <4D808575.90605@liris.cnrs.fr>
To: Joakim Söderberg <joakim.soderberg@ericsson.com>
CC: "public-media-annotation@w3.org" <public-media-annotation@w3.org>
On 03/16/2011 04:33 AM, Joakim Söderberg wrote:
> Dear Pierre-Antoine,
> I can confirm that reading directly at:   http://liris.cnrs.fr/~pchampin/research/mawg/id3/ Works!


> I find your examples [1] very useful. However I'm afraid I don't
> follow what you mean that the RDF produced (from complete.mp3) should
> *entail* the graph complete.ttl graph?

it means that the RDF graph produced by any implementation for 
complete.mp3 should "imply" or "allow to infer" the RDF graph described 
by complete.ttl.

As an example, ext-complete.ttl entails complete.ttl, because it uses 
id3v2: properties instead of the corresponding ma: properties, but it 
adds the sub-property axioms you notices in order to allow you to 
*infer* that the corresponding ma: property holds as well.

> And regarding assigning URIs to id3v2: namespace, I guess you mean:
> id3v2:TCOM     rdfs:subPropertyOf :hasContributor
> id3v2:TEXT     rdfs:subPropertyOf :hasContributor
> etc.
> But the example files don't recommend any particular bindings to
> sub-properties, but only demonstrates one way of how it could be done,
> hence proving that it works,

It only proves that it works if you can perform RDFS inference, or 
*entailment*; if you compare ext-complete.ttl with complete.ttl on a 
purely syntactical basis, you won't be able to recognize that 
ext-complete.ttl is a correct output.

> and we have already stated to the user that
> some semantics may be lost.

Sure; complete.ttl is a correct output (it trivially entails itself) 
although it loses some semantics. It is acceptable to lose some 
semantics, but it is not mandatory (!). So we do not want (I think) to 
rule out ext-complete.ttl as a valid output.

This is why I propose that the test cases verifying RDF output 
explicitly rely on the notion of *entailment*, so as to accept both 
examples as a valid output.

> Or?

Hope that makes it clearer.


> [1] http://www.w3.org/2008/WebVideo/Annotations/wiki/Format_mapping_review
> -----Original Message-----
> From: public-media-annotation-request@w3.org [mailto:public-media-annotation-request@w3.org] On Behalf Of Pierre-Antoine Champin
> Sent: den 9 mars 2011 16:41
> To: public-media-annotation@w3.org
> Subject: mapping, RDF and test-cases
> Hi all,
> [Summary: I explain some issues about RDF-based test-cases, and I
> propose a solution]
> I said in a previous telecon that I had some concern about making
> test-cases for the RDF mapping.
> Let me illustrate my initial concern: when mapping the metadata from
>     http://champin.net/research/mawg/id3/complete.mp3
> I would like the output to look like this [2]
>     http://champin.net/research/mawg/id3/ext-complete.ttl
> However, I can not provide that Turtle file as a reference for test
> cases, because of a number of problems:
> 1/ as suggested by the RDF mapping [1], I minted a number of URIs (in
> the id3v2: namespace) for *subproperties* of the original ma:
> properties, whenever ID3 had a more precise semantics. For example, TEXT
> provides the name of the *lyricist*, so a sub-property of
> ma:hasContributor should be used.
> However, we do not want to prescribe the URIs of such sub-properties, so
> they should not appear in the test-cases.
> 2/ although ID3 does not use URIs for languages, I was able to generate
> a valid URI from the ISO-639 code [2].
> While this is a useful mechanism, it can not be prescribed either, as we
> do not want to sponsor a particular URI-scheme for languages.
> An obvious solution would be to strip the RDF output the bone, stricly
> sticking to what is *required* from the MA ontology. This gives
> something like:
>     http://champin.net/research/mawg/id3/complete.ttl
> which is correct, but loses a lot of information, so nobody would
> actually want to generate *just* this.
> Proposed solution
> =================
> We should provide for the test-cases the stripped version of RDF
> metadata, and we should explicitly state that implementations are not
> required to exactly produce those RDF triples, but to produce something
> from which all those triples can be *inferred* using RDFS semantics.
> In RDF parlance, the RDF produced by an implementation for complete.mp3
> should *entail* the RDF graph expressed by complete.ttl, under the
> entailment regime of RDFS.
> The drawback of this proposal is that it requires that testers use an
> inference engine. However, RDFS semantics is relatively well supported...
>     pa
> [1]
> http://dev.w3.org/2008/video/mediaann/mediaont-1.0/mediaont-1.0.html#correspondence-between-the-informal-ontology-and-the-rdf-representation
> [2] http://lexvo.org/id/iso639-3/eng
Received on Wednesday, 16 March 2011 09:40:43 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 21:17:41 UTC