Re: Action-166 Draft sketch of how to serialize rdf annotation spaces - separate files

On Jul 16, 2008, at 5:11 PM, Peter F. Patel-Schneider wrote:

>> I was worried that was a lot more mechanism, and that given where we
>> were in the process, the desire to get something in to this release  
>> of
>> OWL, and the desire to keep to schedule and move quickly towards last
>> call, the amount of work associated with doing that might have been
>> considered too much. I didn't want to have proposing that much change
>> put the possibility of getting *something* in this space out of  
>> reach.
> More mechanism than using a separate file?  I view using a separate  
> file
> as a *very* heavyweight mechanism.

I'm curious why.

>> In addition there were two technical details which I wasn't sure were
>> covered by the paper. The first is annotations on annotations. I did
>> want to make sure some version of this is possible as I am aware of
>> several use cases in current projects that need it.
> But your proposal only allows a particular kind of very simple
> annotations on annotations.  I don't see how that is adequate.  If you
> think otherwise please put forward the use cases.

All of the cases I've seen SKOS, OBO, OBI, have essentially boiled  
down to provenance information for the annotations - where a synonym  
came from, who supplied a definition, etc. This proposal covers those  
cases. All I've thought I've ruled out are annotations on annotations  
on annotations (and further levels of same). Do you have use cases in  
mind that go beyond the ones the proposal allows? Could you share them?

In the paper the Annotations are fairly limited as well, but could be  
extended. Having a look at the metaontology 
  they have AnnotationByEntity and AnnotationByConstant. Annotations  
are not Axioms and so can't have annotations themselves.

The proposed use of SELF and Annotations (bundles) can do more.

>> The second was the
>> issue of how to view that proposal from an RDF point of view.
> Well, one could just say that the current situation (plus the obvious
> extension to annotations on annotations) is the RDF point of view for
> the treatment in the paper.

We would need more if we were to allow SELF and bundles.

>> I did note that the reification of the axioms and associated  
>> ontology of
>> OWL axioms proposed by the paper seems to not be incompatible with  
>> what
>> I've proposed- in the sense that I can see how one could script the
>> creation of the reified axioms, and then import that into the second
>> file.
> Surely it is easier to do this without having to invoke a special- 
> purpose file-handling mechanism.

The file was intended to be OWL-DL, so that the exact same file  
handling that would apply to O would apply to m(O). The sole  
restriction was that ontologies that were imported into it shouldn't  
have annotations themselves, so as to avoid generation of further  
files.  A (perceived) advantage of having the separate file was that  
there was less need for multiple levels of reification to keep the  
annotation facts out of the domain (in the case of bundles).  
Correspondingly, if there is a single file, there is more mechanism  
needed to extract out the two ontologies for reasoning separately.

However, one thing I hadn't thought through was interaction with  
imports. Quickthink suggest that if O imports O', m(O) would need to  
import m(O'), but I've added this to the known issues.

>> However, I like that paper and its approach and if the WG were  
>> wiling to
>> seriously consider what it proposes I'd certainly support that.
> Me too.
> Here is a quick summary of what I see as the gist of the proposal in  
> the
> paper.  I'm leaving out a few details.
> Regular reasoning on an ontology proceeds as normal in OWL 2 DL.
> Annotation reasoning on an ontology O is done by constructing a
> meta-ontology m(O) containing facts corresponding to the annotations  
> in
> O plus facts corresponding to the structure of the axioms in O.
> One could envision a stripped-down version that doesn't include all of
> the last bit - only individuals corresponding to each axiom in O.
> It seems to me that the two-file solution has all the machinery of  
> this
> stripped-down version, plus the complications involved with having two
> files.

That's my take as well, subject to the question about where the  
complications arise. It seems that it is a tradeoff of where the  
mechanism is - more files or more specialized encoding/decoding (which  
might complicate use in OWL-R). Solely based on the services  
described, the two file approach might be more appealing to some  
because of the lack of reifications on reifications in the RDF, and  
the implemented (if not standard) means for working with named graphs.

Something else to consider is whether a two file solution for RDF  
necessitates that the XML syntax do the same thing. I would hope that  
if we did land up with two files in RDF, we could still leave the XML  
version in a single file but haven't thought through how this would  
work. I've added this to the known issues as well.

What I found appealing about the paper is the part that goes further -  
the definition of the metaontology, the clearer statement of  
semantics, and the query language.


> There are some "extras" that would be useful to have in this proposal,
> including the ability to put arbitrary axioms into m(O).

Received on Thursday, 17 July 2008 06:38:35 UTC