Re: proposal - Fragments redux (unifying the threads under Issues 75-80)

On Nov 30, 2007, at 5:28 PM, Ian Horrocks wrote:

> On 29 Nov 2007, at 17:06, Jim Hendler wrote:
>
>>
>>
>>
>> On Nov 29, 2007, at 2:44, Carsten Lutz <clu@tcs.inf.tu-dresden.de>  
>> wrote:
>>
>>> On Wed, 28 Nov 2007, Jim Hendler wrote:
>>>>
>>>> well, it's not so much motivated by computational properties,  
>>>> see out in the real world there's people who just implement fast  
>>>> engines and don't worry so much about the details...
>>>
>>> Sorry to object, but IHMO this approach is precisely why the  
>>> original
>>> OWL Lite was broken. And I understood we wanted to fix this?! We  
>>> should
>>> at least understand the computational properties of the fragments we
>>> are selecting.
>>>
>>
>> IMO, its because we worried too much about theory that lite is broken
>
> I can't let this go by without a correction. OWL Lite is broken  
> because it was designed without *any* consideration of theory --  
> the design "methodology" was to haggle over what features to throw  
> out and what to leave in, without any (rigorous) analysis of the  
> (computational) consequences. There was even a last minute push to  
> have oneOf be included on the grounds that it is a "must have"  
> feature for many users; this was only given up when it was pointed  
> out that the resulting language would have exactly the same worst  
> case complexity as OWL DL.
>

but Ian, this is just the point - nominals by themselves are no  
problem (I think even EL++ allows the equivalent of hasValue - i.e.  
OneOf with a single value) and many DB and Rule languages have them.   
We added too many things and then cut back based on DL reasoning  
issues (what I meant by the theoretical in the above).  So we never  
really looked at starting from scratch with language design issues  
involved and people who have, like HP and  Oracle (I use these  
because they are in the WG, there are others as well) have developed  
other subsets.  In a number of papers, as unnamed fragment centered  
somewhere near the thing I called RDFS 3.0 has been identified as the  
most used, and it has been implemented by a much larger number of  
implementors than have implemented full OWL DL tools - i.e. all the  
various FOAF tool providers, several of the Web 3.0 folks, several  
open source triple stores, and various of the academic systems demoed  
at ISWC.
  btw, this subset is not designed without consideration of  
theoretical concerns, this subset can be supported by datalog with  
some fairly simple restrictions
  -JH

Received on Monday, 3 December 2007 14:45:41 UTC