Re: Distinguishing fragments

I think the answer is pretty clear:

OWL-DB aka DL-Lite works well for people who have or want to use data  
in (possibly existing) RDBs and who want to be able to map  
(conjunctive) queries against the ontology to SQL queries against the  
DBs. But they need to be able to live with a very restricted subset  
of OWL to make this possible.

As Achille pointed out, OWL-R DL will work for people who can't live  
within the expressive constraints of DL-Lite and/or who want to use  
Datalog/LP technology in their implementations. (It might also help  
to establish a link with RIF.) And as Zhe pointed out, it might be  
useful as a restricted version of OWL-R Full

Clearly we need to figure out a suitable naming scheme and to add  
some introductory material (including, in particular, the rationale  
for each fragment), but this is only to be expected.

Ian

p.s. are these folks from the RDF community happy with the above  
explanation and/or willing to talk to us directly?


On 7 Mar 2008, at 19:42, Jim Hendler wrote:

> Actually, the first message I sent was that I got some folks from  
> the RDF community who surprised me because they didn't like our  
> current fragments document, which I thought was pretty good.  THe  
> primary questions they asked were as to what the reason was to have  
> two DL fragments that seemed similar in many ways (OWL R DL and DL- 
> Lite) and they suggested we drop one.  They also asked, if the idea  
> was symmetry, why we would not have Full and DL versions of all the  
> fragments - if it is useful to the "reasoning" community to have DL  
> segments of these, why would it not be useful for the "RDF"  
> community to have Full versions of the others -- after all, all of  
> those would increase interoperability between the OWL and RDF  
> worlds (their words, not mine).  I raised this off list with those  
> who had volunteered to be on the fragments TF because I didn't want  
> it to look like it was an attack on the document (which, again, I  
> thought was pretty good).  However, as at least a couple of the  
> people I talked to were on the AC of the W3C, I thought it might be  
> worth anticipating their comments and working them through.
>  After several messages, which Ian has summed up below, modulo some  
> minor changes (See inline)
>
>
> On Mar 7, 2008, at 12:30 PM, Ian Horrocks wrote:
>
>>
>> We wanted to move a discussion on distinguishing fragments to the  
>> list. Let me start by summarising what has been said so far (I  
>> used my best endeavours, but others may want to add corrections/ 
>> clarifications):
>>
>> Jim: why do we need a DL version of OWL-R? Couldn't it be  
>> confusing with DL-Lite?
>>
>> Boris: OWL-R allows for moving between RDF and DL worlds -- may be  
>> important to some.
>>
>> Jim: people I talked to said they would only use DL-Lite xor OWL-R
>      this is DL-Lite XOR OWL-R DL (to clarify)
>>
>>
>> Achille: Agrees that there is potential for confusion especially  
>> due to fragment names; OWL-R DL covers datalog reasoning and  
>> colleagues in China have a need for it.
>>
>> Boris: suggests OWL-T (taxonomic), OWL-R and OWL DB for EL++, DLP/ 
>> rules and DL-Lite fragments.
>>
>> Jim: we would end up with the following naming scheme:
>>
>> OWL Full and OWL DL
>> (OWL lite) - we hope to deprecate
>> OWL-T DL
>> OWL-T Full
>> OWL-DB DL
>> OWL-DB Full
>> OWL-R DL
>> OWL-R Full
>>
>> How would a user choose between OWL DB DL and OWL R DL? Thinks  
>> that OWL-R and OWL-DB are both motivated by large scale data rich  
>> applications, so would have a problem with the name OWL-DB for one  
>> and not the other. Suggests OWL Datatlog and OWL Rules, but  
>> overlap issue more important than names.
>>
>> Zhe: OWL-DB is confusing; Oracle has no short term plan to  
>> support; suggests OWL-A (A for Abox) or OWL-I (I for instances)  
>> instead of OWL-DB. Thinks OWL-R DL might be useful as a restricted  
>> version of OWL-R Full.
>>
>> Jim: RDF triple store folks said they were confused about when to  
>> use DL-Lite and when to use OWL-R DL. Suggest that it may be OK to  
>> have asymmetric fragments (not always both Full and DL versions).
>>
>> Boris: OWL-R Full is for RDF users wanting more expressive power;  
>> OWL-R DL is for OWL 1.1 DL users wanting to down-size. Nice thing  
>> is that both communities can meet at the same level.
>>
>> Ian: Don't need symmetric fragments -- omit those with no clear  
>> rationale (such as OWL-DB Full). Choosing between OWL-DB and OWL-R  
>> should be easy: if you can live within the restrictions imposed by  
>> OWL-DB and/or want to use data directly from relational DBs, then  
>> use OWL-DB; otherwise, use OWL-R. Agrees that design issues more  
>> important than naming.
>>
>> Ian: Can appeal to large community when they realise that they can  
>> use OWL-DB with existing relational DBs. This isn't to detract  
>> from OWL-R. Agrees with Achille that OWL-R DL can be be useful for  
>> LP implementers.
>>
>> Ian: Need better names and a clear rationale (in fragments intro).  
>> Case for "OWL-DB" is a no-brainer -- can use OWL directly with  
>> data that is already sitting in relational DBs.
>>
>> Jim: asymmetry favours DL over Full?
>>
>> Jim: but why would we recommend to such users that they use OWL R  
>> DL instead of DL-Lite?  Also if this is true of OWL R DL, why not  
>> DL-Lite Full? Shall we move discussion to list?
>>
>> Ian: asymmetry means only including fragments for which there is  
>> clear rationale.
>>
>> Ian: If users can't live within the restrictions imposed by DL- 
>> Lite (it is quite a restricted language), then they should use OWL- 
>> R. Yes, let's move to list.
>>
>>
>
> "If we knew what we were doing, it wouldn't be called research,  
> would it?." - Albert Einstein
>
> Prof James Hendler				http://www.cs.rpi.edu/~hendler
> Tetherless World Constellation Chair
> Computer Science Dept
> Rensselaer Polytechnic Institute, Troy NY 12180
>
>
>
>

Received on Friday, 7 March 2008 20:26:14 UTC