Re: [model] Why Motivations cannot be on Bodies

> On 18 Jun 2015, at 18:50 , Robert Sanderson <azaroth42@gmail.com> wrote:
> 
> 
> Apologies in advance as this likely reads as lecturing, but there's a lot of background that needs to be understood (or at least accepted) to explain why motivations cannot be directly associated with bodies.

With the same apologies, let me add another aspect to describe the problem.

The problem is that additional features, like motivations, are, conceptually, meant to describe the *association* of the annotation and the body and not the body itself. Ie, if we annotate something with the picture of Paris, what we *really* want, conceptually, is to say that the *association*

 <annotation> body <picture of Paris>

has been made for specific motivation. Using RDF terminology, we want to set a predicate against the triple *as a whole*, as if it was one resource.

The difficulty to do that in RDF may have created more discussion that anything else in RDF land. In the current model, when we use "motivatedBy" on an annotation, we are defining, at least in my view, a (benign) hack by setting the motivation on the <annotation> itself: because there is only one body, doing this is absolutely o.k and it works. However, it breaks down as soon as there are several bodies that have different motivations.

The 'specific resource' trick is a way around this problem, again using a benign conceptual hack. The Provenance vocabulary had a different solution (the 'qualified' patterns, ie, for each property there is a 'qualified' counterpart for that purpose[1], which would be way to heavy for our goals). Outside of RDF the concept of property graphs (used in some NonSQL databases) has great facility doing exactly that, but property graphs are still a moving target, not stable enough for being used in a standard and, besides, it would make us loose all connections with Linked Data. Finally, there was a brief discussion on using named graphs earlier; systematic usage of named graphs would be yet another alternative (afaik, JSON-LD is pretty o.k. using named graphs, so maybe we should look into this if we really want it…), but one must wonder whether it is worth the trouble.

I hope I did not muddy the waters further…

Ivan


[1] http://www.w3.org/TR/prov-primer/


> 
> Linked Data and RDF assume a globally scoped, open world model, in which anyone can make any assertion about any resource, and the context of that assertion is not at all relevant.  Thus if I make the assertion that a photo depicts Paris, and someone else makes an assertion that the same photo depicts Tokyo (they both have big towers after all), that's completely acceptable.  If this wasn't the case, then there would be semantic battles in the same way as wikipedia has edit wars, but in a fully distributed system.  Further, if something is not asserted, then it is assumed to be unknown, not false.  Just because I've asserted that the photo depicts Paris doesn't make it wrong, a priori, that it also depicts Tokyo (it could be a montage).  It would also depict many other things beyond Paris even if it didn't depict Tokyo.
> 
> The side effect of this is that when you want to make explicit claims about resources that are only true in a particular context, you cannot make the claim about the resource in general.  As bodies (and targets) of Annotations are web resources, not properties of the Annotation, if we want to make claims about that resource in the context of the annotation, we need a new resource ... which is a Specific Resource.
> 
> We already have this in the model to enable Selectors, States, Styles and Scope.  You cannot make an assertion that an image has a segment described by a Selector, as many annotators might annotate different segments of the same image.  In the open world, all of those selectors are possible and true ... so when looking at the graph you cannot distinguish which selector belongs to which annotation.
> 
> Thus the Specific Resource stands for the Body or Target in the specific context of the Annotation.  We can safely make assertions about the Specific Resource, as we just created it for that purpose, without colliding with other annotations' assertions.
> 
> Given that bodies have the same required features as targets, we have Specific Resources for bodies.  For example, we could otherwise not say "this part of this web page is about that part of that image".  Or this blog at this point in time is about that news article.  This part of this video is about that part of that data set. And so forth.
> 
> Thus, and there will be a test on this next class, we cannot make annotation specific assertions about either the body or the target directly on the resource and motivation is (clearly) an annotation specific assertion, not a global one. We already have a solution for that, which is the specific resource construction.  This leads to the proposal to allow motivatedBy on specific resources. (Or potentially and more radically, to an alternative to do away with multiple bodies completely and have annotation sets as first class objects)
> 
> 
> Thanks and hope that helps!
> 
> Rob
> 
> 
> --
> Rob Sanderson
> Information Standards Advocate
> Digital Library Systems and Services
> Stanford, CA 94305


----
Ivan Herman, W3C
Digital Publishing Activity Lead
Home: http://www.w3.org/People/Ivan/
mobile: +31-641044153
ORCID ID: http://orcid.org/0000-0003-0782-2704

Received on Saturday, 20 June 2015 06:03:36 UTC