Re: ISSUE-67 (reification): REPORTED: use of reification in mapping rules is unwise

Ian Horrocks wrote:
> On 21 Nov 2007, at 11:59, Jeremy Carroll wrote:
> 
>>
>> Boris Motik wrote:
>>> Hello,
>>> I don't want to get into an argument here whether this is really 
>>> needed or not; however, I wanted to point out that I spoke to quite
>>> a few people asking for the annotation of axioms.
>>
>> There are always trade offs. I certainly am not trying to suggest that 
>> the feature was unmotivated. But *need* is quite a strong word.
>>
>> In the wider scheme of things, the Web and Semantic Web are not 
>> *needed*, (not in the way that we need bread and water; love and 
>> friends) so I am pretty sure that any feature of OWL 1.1 is not 
>> necessary either.
> 
> Come on Jeremy -- this is a rather silly argument. 

It might be silly, but I am still feeling uninformed about what the 
annotation system is actually meant to do.

> Of course OWL (1.1) 
> and the Semantic Web "don't amount to a hill of beans in this crazy 
> mixed up world", but this is hardly pertinent, and I think that we all 
> know roughly what is meant in this context by "need". Just to make this 
> a little more concrete, it is clear that improving/extending the OWL 1.0 
> annotation framework (or lack of it) is a key requirement for and 
> motivation of OWL 1.1 (see http://www.w3.org/2007/OWL/wiki/Overview). As 
> Alan has already pointed out, this is supported by conversations with 
> experienced users representing key user groups; for example, one of the 
> main reasons given by the OBO community for *not* using OWL is its very 
> weak annotation mechanism.

The OWL 1.0 annotation system appears to me to be very rich:
   annotations are typed (by the annotation property),
   annotations can be simple strings or they can be complex objects,
   the annotation system is arbitrarily extensible,
   in OWL Full:
    + the annotation have an extensible hierarchy (from subPropertyOf)
    + and annotations can be declared as being dates or integers or 
whatever.

This strikes me as an incredibly rich annotation system; richer than any 
web language that I am aware of; richer than say javadoc; richer than 
CVS tags; richer than however you would go about annotating an RDBMS.

The OWL 1.0 method does have some issues:
- annotations are attached to specific items in the KB
- annotations have semantics that may interact with the the semantics of 
the KB

As far as I understand, these two aspects are sufficiently problematic, 
that we will spend a significant amount of our effort on annotations.
It is not unreasonable to want to unpack what the requirements are, and 
to be suspicious of unsubstantiated statements of need.


> 
>>
>> More constructively, what I am hearing, I think, is that the 
>> requirement is for comments that have no semantics and just fit into 
>> the specification in the right way.
>>
>> In RDF/XML there has always been the capability to have such comments 
>> - they look like:
>>
>> <!--
>>  This is a comment, it has no bearing on the formal semantics of the 
>> document.
>> -->
>>
>> It may be possible to provide say, an informative GRDDL transform from 
>> an XML version of the axioms, to RDF/XML, and back again, that round 
>> trips comments appropriately.
> 
> As you probably recall, this was discussed and dismissed in the WebOnt 
> working group for the reason that it is much too low-level (it is a 
> feature of XML), and that such comments may be lost when documents are 
> processed. 

It depends on the use cases ...

I was partly surfacing it to test whether the view is that RDF/XML is 
still the exchange language of choice. XML comments within RDF/XML 
annotating OWL ontologies would be difficult to get working 
interoperably. But then Bijan's solution doesn't feel like an RDF/XML 
solution either.



> What you propose w.r.t. GRDDL sounds more like a hack than a 
> realistic solution. Moreover, this mechanism would hardly satisfy the 
> requirement to have a more comprehensive framework that allowed (at 
> least) for annotating axioms as well as entities.
> 

The point of suggesting them is that because they can go (almost) 
anywhere in an XML document, there could be conventions that permitted 
annotating axioms as well as entities.

But if the point is that what is being annotated is a triple store, then 
they won't work - but without describing use cases clearly it is hard to 
know what the requirements are.

Jeremy

Received on Monday, 26 November 2007 15:30:04 UTC