Re: Annotations and the Graph

On Tue, Oct 27, 2015 at 3:49 PM Jacob Jett <jjett2@illinois.edu> wrote:

> Hi Randall,
>
> Can you explain more about the use case for this?
>

Sure.


>
> I think quite possibly that this proposal might result in a model that
> breaks reasoners based on OWL. I remember reification being discussed at
> one of the early OAC meetings and as I recall the models avoidance of
> reification and scoping the relationship between the body and target is on
> purpose. Perhaps Rob or Tim might recall why this was avoided but its not
> clear to me if your proposal does not in fact make the standard unusable
> for anyone employing reasoners, more or less defeating its semantic web
> purposes.
>
> It blurs the line between semantics specific to the annotation model and
> those specific to RDF.
>
>
I thought I was pretty clear about the use case, but I'll try again.

The use case is representing and querying annotations in a consistent and
straightforward way.

If a publisher posts an article (schema:Article) and its comments
(schema:Comment) it relates them with a simple predicate (schema:comment).

As an independent reader who wants to make my own comments, not necessarily
involving the commenting system that the publisher has blessed, I want to
use the same pattern.

I want to take N repositories and query them for (Article, comment, ?) and
get back all those objects which are comments on the article.

Contrary to your concern, I was actually under the impression that my
proposal makes things easier, not harder, for reasoners.

Unless of course you are using "easier" to mean it's easier to ignore
something than to try to derive meaning from it.

One would think, I hope, that an oa:commenting annotation means that the
body is a comment on the target.

In order to reason that out, one needs to express that <the property chain
that traverses the inverse of hasBody, through an an Annotation resource
with the motivation oa:commenting, to the object of its hasTarget> implies
a <comment> relationship between the two.

The only complication my proposal presents to reasoners is that *only if
they want to instantiate the triples described by the annotation* they will
have to either namespace (named graphs, quads) the triples or deal with the
fact that they are reasoning on this user data, which may be conflicting or
inconsistent if it's not validated and trusted.

That's data life, though. Validate and trust or quarantine and contain.

Hoisting the predicate onto the annotation as a motivation seems to be
serving the same purpose as RDFS reification: express the whole triple as a
resource that can carry metadata, such as additional provenance
information. That's similar also to use cases for a named graph.

It could be that I'm misunderstanding something fundamental about named
graphs or RDF. I'm especially curious to hear whether there are reasons why
RDFS reification was considered and rejected.