Re: [model] Why Motivations cannot be on Bodies

Hi Doug,

On Thu, Jun 18, 2015 at 10:32 AM, Doug Schepers <schepers@w3.org> wrote:

> In particular, I have difficulty justifying the assertion that the bodies
> are by necessity global resources, not local to the annotation.


Because that's one of the fundamental requirements of RDF and Linked Data.
Every triple is asserted with a global scope, regardless of the document or
system that any client may have discovered it in.  Triple stores simply put
all of their triples into one big pool, leading to some folk to refer to
"the big triplestore in the sky" that theoretically (and implausibly)
contains all triples asserted anywhere.

The body of an annotation is a web resource, not an exclusive property [in
either sense] of a single annotation. Thus the need for specific resources
to express things are are only true of the annotation's use of the body
resource.


> Outside the context of the annotation, you lose all other context,
> including the target and the provenance. So, how is a global body without
> context a useful statement?


A body can be used in many different situations and contexts. Some of which
might be annotations, some might be elsewhere.  For example, a youtube
video about something would be a very reasonable body of an annotation that
formally linked it to the resource that it describes.  You're saying that
the youtube video is not useful without the little bit of JSON that links
it to something else?


> And even granting that it is theoretically useful, if the annotation data
> model is made rather more complicated by including this concept, is that a
> compromise worth making in the data model?
>

Yes, in my opinion. The alternative is to throw out all of the web
architecture and the notion of linked data for something that can easily be
solved in several different ways that would be completely compatible with
existing work, and already exist within the model.


> I also didn't understand your claim that you can't make an assertion about
> a segment of an image, just because other assertions can be made about it.
>

The segment is only relevant within the context of the annotation, not
globally.

Three annotations, each of which is about a different part of the same
image would thus be:

anno1 hasTarget image
anno2 hasTarget image
anno3 hasTarget image
image hasSegment "100,100,640,480"
image hasSegment "0,0,500,500"
image hasSegment "100,0,400,200"

Which segment goes with which annotation?  It's impossible to know, and
that would be the situation without the Specific Resource.

Similarly:

anno1 hasBody video
anno2 hasBody video
anno3 hasBody video
video motivatedBy commenting
video motivatedBy commenting
video motivatedBy replacing

Which of the annotations are the two comments, and which is the one where
it should replace the target?
Same problem.

Now consider ordering.  Because a resource can be in multiple lists at the
same time, you need to have some intermediary similar to our specific
resource.  This necessitated the rdf:List construction, for a simple three
item list of [item1, item2, item3]:

list first item1
list rest list2
list2 first item2
list2 rest list3
list3 first item3
list3 rest nil

Or Proxies in the ORE ontology, ListItems in Collections Ontology, Slots in
the Ordered List Ontology, and so on.
In comparison, the multipurpose specific resource construction is cheap and
understandable :)

Hope that better explains the situation.

Rob

-- 
Rob Sanderson
Information Standards Advocate
Digital Library Systems and Services
Stanford, CA 94305