Re: CR Exit Criteria

Ian et al --
   I think the new exit criteria [1] are better, but I have two minor  
concerns  (which actually are for the opposite reason from before --  
I've now been convinced by you, Bijan and others that there are  
implementations out there that are likely to pass the tests and to be  
able to demonstrate all the features I've been questioning ), I think  
some of the wording in the current version could be tightened up -  
because the way things are said now there's a bit too much room for  
interpretation and I think we would be subject to being made to  
tighten things up and do a longer CR than we have to if someone wants  
to cause delay or complain.  In particular, I suggest a couple of  
wording changes:

I) The current version [1] says for OWL Full

> At least two different implementations of an OWL 2 Full entailment  
> checker implementing useful subsets of OWL Full and passing a useful  
> subset of the non-DL test cases


but what is a useful subset? Rather, since we know OWL 1 Full  
implementations exist (proved when OWL 1 passed CR), and since those  
would by default handle a useful subset of OWL 2, I don't think this  
CR needs to prove the implementability of Full.  Rather, we should  
just show that the new features added in OWL 2 Full are implementable  
- so let's only test the new stuff.  So I suggest that we make the  
only OWL 2 Full criterion:

  At least one implementation passing each of the non-DL test cases  
that support features of the OWL 2 Full design that were not in OWL 1.


II) The current version [1] says for each of the Profiles (EL, QL and  
RL)

> Two different implementations of an OWL 2 XX entailment checker, at  
> least one of which takes advantage of some of the claimed benefits  
> of the XX profile

again, "some" makes me nervous, and also some of the claimed benefits  
of these profiles are theoretical, not implementable, and thus an  
implementation couldn't show them.  Rather, I'd suggest we say

> Two different implementations of an OWL 2 XXX entailment checker and  
> at least one implementation that can demonstrates each of the  
> (implementable) benefits of the XXX profile


that will be more easily demonstrated and the only thing that would be  
debatable is which things are implementable, but I think those won't  
be too hard for the WG to decide - and we'd be able to decide during CR.

I think these criteria would be clearer than those in [1] at the time  
of this writing, and also that it would be easier to prove that all CR  
criteria are met than with the less specific ones.  I also think that  
we are very close to the above already, and that the criteria being  
proposed would thus be easily met.

   -Jim H.












On May 23, 2009, at 1:47 PM, Ian Horrocks wrote:

> Yet another re-working of the exit criteria is now available in the  
> usual place [1]. As well as adding the improved "at risk" text, it  
> also separates the Full, DL and Profiles cases a little more  
> clearly. For Full I now have:
>
> "At least two different implementations of an OWL 2 Full entailment  
> checker implementing useful subsets of OWL Full and passing a useful  
> subset of the non-DL test cases. "
>
> No doubt you will let me know if you don't like this formulation.
>
> Ian
>
> [1] http://www.w3.org/2007/OWL/wiki/CR_Exit_Criteria
>
>
> On 23 May 2009, at 04:33, Sandro Hawke wrote:
>
>>
>>> On Fri, May 22, 2009 at 7:36 AM, Peter F.Patel-Schneider <
>>> pfps@research.bell-labs.com> wrote:
>>>
>>>> From: Ian Horrocks <ian.horrocks@comlab.ox.ac.uk>
>>>> Subject: Re: CR Exit Criteria
>>>> Date: Fri, 22 May 2009 06:04:44 -0500
>>>>
>>>>> On 21 May 2009, at 00:53, Peter F.Patel-Schneider wrote:
>>>>
>>>> [...]
>>>>
>>>>>> 1. Resolve dependencies on rdf:text (currently at Last Call)  
>>>>>> and XSD
>>>>>>   1.1 Datatypes (currently at Candidate Recommendation).
>>>>>>> or put rdf:text as risk as described in previous email
>>>>>
>>>>> We agreed to make it "at risk" in the spec, so we should mention  
>>>>> it
>>>>> here. I added "Note that rdf:text is marked as "at risk", and  
>>>>> may be
>>>>> removed from the OWL 2 specification."
>>>>
>>>> Umm, if rdf:text goes away OWL 2 needs something to replace it  
>>>> with, so
>>>> I think that the wording should be something like
>>>>
>>>> "Note that rdf:text is marked as "at risk", and may be replaced  
>>>> with a
>>>> datatype with name owl:text that serves the same purpose.
>>>> Implementations will be able to switch to owl:text by simply  
>>>> replacing
>>>> occurences of rdf:text with owl:text."
>>>
>>>
>>> There are other alternatives. At the moment, for instance, there is
>>> discussion of using the name rdf:plainLiteral. I suggest something  
>>> more
>>> along the lines:
>>>
>>> "Note that rdf:text is marked as "at risk", and may be renamed,  
>>> replaced, or
>>> have technical details modified as a result of ongoing work of the  
>>> joint
>>> OWL/RIF working group determining its specification."
>>
>> +1  I was going to say that.
>>
>>      -- Sandro
>
>

"Con un poco de semántica ya se consigue ir muy lejos"

Prof James Hendler				http://www.cs.rpi.edu/~hendler, @jahendler,  
twitter
Tetherless World Constellation Chair
Computer Science Dept
Rensselaer Polytechnic Institute, Troy NY 12180