Re: concerns about RDF Mapping doc

Thanks for this elaboration! I look forward to further details.

On 18 Oct 2007, at 15:34, Jeremy Carroll wrote:

> At the telecon last night, I (successfully) requested a weakening  
> of the proposal for first publication, to leave it open as to  
> whether the RDF Mapping doc would be included or not.
>
> This was not an attempt to immediately advocate that it should not  
> be included; merely an attempt to leave it open to advocate that it  
> should not be included.

Sure, but then this interacts with my current push to get things  
published sooner rather than later. In particular, in an October time  
frame. I'd rather publish some things sooner than publish all three  
together, so I am open to this weakening, but regard it as  
unnecessary and undesirable.

> My motivation for doing this was that in an HP internal reviews of  
> all three documents, the reviewer had minor issues with the other  
> two docs but had more major issues with RDF Mappping.
>
> Before advocacy one way or another, I will need to refresh my  
> memory: looking at both the most recent version of RDF Mapping, and  
> the HP internal review.
>
> If I remember correctly the heart of the criticism was that:
>
> a) The OWL 1.1 suite of docs, do not include an OWL Full semantics.  
> HP regard an OWL Full semantics as critical.
>
> b) In OWL 1.0, the tension between the OWL Full semantics and the  
> direct semantics is resolved in cleverly non-deterministic mapping  
> rules

This is the first time I've heard this view. It never occurred to me  
as a possibility.  Interesting. I'll think about it.

> c) The OWL 1.1 mapping rules, probably in order to simplify life  
> for the implementers, have reduced the non-determinism (and the  
> cleverness)

In part. Also, I think it helps users if there is a closer  
correspondence. In my experience with serializations in Swoop, having  
a nice mapping was important and have a predictable mapping was  
welcome. There was some comments on this on public-owl-dev recently.  
I would even support having a deterministic serialization to RDF/XML,  
as an option, specced by this group. Ideally, it would look very much  
like a stricter version of an XML serialization, with class axioms,  
property axioms, and facts grouped. Both Swoop and Protege4 have  
reasonably structured serializations.

> d) In the HP reviewer's opinion the reduction in the non- 
> determinism severely impacted the ability to define an appropriate  
> OWL 1.1 Full semantics.
>
> In my view, if this criticism stands, then, it is a criticism of  
> the whole document, rather than of a few paragraps or sections.

Doesn't this depend on whether the non-determinism can be recapture  
by, for example, adding optional parts to various productions? Then a  
non-deterministic mapping could be built as an extension of the  
deterministic one and if you want the deterministic one you just  
ignore all the optionals.

> It also reflects a value judgement that OWL Full semantic  
> compatibility is more important than simplifying life for the  
> implementors of the mapping rules, even at the risk of being too  
> clever by half.
>
> Jeremy
>
> (carefully sketching a position that I might hold later, or some  
> anonymous HP reviewer might hold; without actually committing to it!)

As I said, I'll be interested in the fleshed out version.

Cheers,
Bijan.

Received on Thursday, 18 October 2007 17:32:44 UTC