Re: flat vs. structure, and embedding

Hello Pierre-Antoine,



Pierre-Antoine Champin さんは書きました:
> Hi all,
>
> the recent discussion about formats / structure made me think about
> something.
>
> It seems to me that
> - the arguments in favour of a structured model are semantics related
> (mainly to satisfy more requirements)
> - the arguments in favour of a flat model are syntax related (mainly to
> allow embedding of metadata in media objects) or API related (ease of use).
>
> So, wouldn't it keep the best of both worlds to provide:
> - a structured conceptual model (e.g. an RDFS or OWL ontology)
> - a simple (ideally flat) syntax that would allow to represent the most
> common patterns of that conceptual model.
>   

We have not yet decided whether we need to develop our own format, no 
matter if structured or flat. IMO our main focus should be on 
interoperability between existing formats, without defining a simple or 
complex new one. See also my prevoius mail in the "3 o more layer" 
structure thread.

To give a concrete example: your proposal

md = file1.getMetadata()
 title = md.getProperty("title")

could simply use the media file itself as "file1" and be

md = file1.getMetadata()
 title = md.getTitle


if we do not define our own format. Of course, this does not implement a 
requirement - having extrinsic metadata.
http://www.w3.org/Bugs/Public/show_bug.cgi?id=6109
But I think this is not a central requirement for version 1.0 of our work.


A comment on a part of your proposal:

"The notion of inheritance, on the other hand, should probably be 
specified by the ontology. Especially, what should happen when different 
values are inherited (through different paths) for the same property 
(here: creator)? Should the "closest" one have precedence ? Should they 
be merged? "


These are questions we have to answer if we go for the "have a format 
for the ontology" approach. In your case, it is a generell graph 
structure, not specific to RDF or XML. Compare this to the precedence 
description at
http://dev.w3.org/2008/video/mediaann/mediaont-api-1.0/mediaont-api-1.0.html#property-createDate
It contains statements like
"exif:DateTimeDigitized (based on [MWG Guidelines Image]). This is taken 
only into account if both xmp:createDate and exif:DateTimeOriginal are 
missing."
IMO this propose is clear enough for implementers and has the benefit 
that they (and we) do not need to answer the questions you have posted.

Felix


> Normalized server-side metadata could use the richer language
> internally, while embedded metadata would use the simple syntax --
> possibly loosing some information, but beeing consistent with the
> complete metadata.
>
> I put some thoughts about the subject on the wiki.
>
> http://www.w3.org/2008/WebVideo/Annotations/wiki/Flat_syntax_for_structured_model
>
> What do you think?
>
>   pa
>
>
>
>

Received on Saturday, 22 November 2008 02:14:16 UTC