RE: Undiscussed DL-Lite Changes: SameIndividuals out, UNA + Functional Properties in


Please allow me to recapitulate a few technical facts about DL-Lite. If one takes FOL-reducibility of query answering as the guiding
principle for the design of the fragment, then one unfortunately must live with the fact that any kind of reasoning about equality
must be disallowed in DL-Lite. Hence, you can have neither functional properties nor keys (which can actually be thought of as a
kind of fancier functional properties). Note that the fact that the keys are easy doesn't help; in fact, query answering is not
FOL-reducible already in the case when the only thing you have is the ABox and just one inverse property axiom.

Please allow me also to clarify the design of DL-Lite in OWL 1.1, which allowed for SameIndividual, but disallowed
(inverse-)functional properties. The reason why, in that case, you can have SameIndividual is because these axioms do not require
true equality reasoning: you can deal with them such that, before reasoning, you simply replace one individual with the other one.
However, as soon as you add any kind of "true" equality reasoning, you need to assume UNA as well if you want to stay FOL-reducible;
well then, the only logical thing to do was to throw out SameIndividual.
The current design of DL-Lite thus pushes the keys to the limit of what is theoretically possible: you can use key axioms as a kind
of constraints that check whether the data in the ABox has been properly structured. (In my previous e-mail, I said that easy keys
don't do anything in the current variant of DL-Lite, which was incorrect: while keys cannot entail new equalities, they can be used
as checks.) Thus, I'm afraid it is theoretically impossible to "get more for your money" in DL-Lite without violating the
fundamental design principle of the fragment.

I would also like to mention that I changed the Profiles document only after talking to Diego Calvanese (at the DL Workshop in May
2008), who has advised me on the maximal set of constructs that one can have in the fragment and who has checked my changes.



> -----Original Message-----
> From: [] On Behalf Of Ian Horrocks
> Sent: 10 July 2008 15:01
> To:
> Subject: Re: Undiscussed DL-Lite Changes: SameIndividuals out, UNA + Functional Properties in
> DL-Lite doesn't have UNA as such -- but the definition says that to
> be in the DL-Lite fragment an ontology must include
> DifferentIndividual statements for all individuals in the ontology.
> This means that the fragment is still a fragment (every DL-Lite
> ontology is a DL ontology and has exactly the same semantics in DL).
> The DifferentIndividual statements effectively *assert* UNA. This is
> needed if the OWL DL-Lite profile is to enjoy the computational
> properties that are its "Raison d'etre" (because the profile also
> includes functional roles). Having same individual statements in DL-
> Lite would thus be rather pointless, as asserting SameIndividualAs
> for any two non-equal individuals would result in ontology
> inconsistency.
> Regarding easy keys, it isn't quite true to say that they "don't do
> anything". In the presence of (asserted) UNA they behave rather like
> database constraints -- two individuals having the same key is a
> constraint violation and would again lead to an inconsistency.
> Ian
> On 7 Jul 2008, at 18:05, Michael Smith wrote:
> >
> > In the last telecon, during discussion of easy keys [1], it was stated
> > that UNA applied to the DL-Lite profile, which I questioned.  This was
> > not the case at F2F2, so I referenced the document history.
> >
> > In a series of edits to the profile doc on 2008-05-14, SameIndividual
> > axioms were removed from the DL-Lite profile, and the UNA was added
> > along with functional property axioms.  I'd like these changes to be
> > discussed by the WG, perhaps with comment on why they were initially
> > made without discussion.
> >
> > My opinion -- influenced largely by discussions at F2F2 -- is that (1)
> > SameIndividual axioms are a more desirable feature than functional
> > properties and (2) a profile of OWL with the UNA is problematic.
> >
> > This conversation is relevant now because (I believe) the
> > applicability
> > of easy keys to DL-Lite is dependent on the UNA.
> >
> > Thanks,
> > --
> > Mike Smith
> >
> > Clark & Parsia
> >
> > [1]
> > 2008-07-02#action_165_Investigate_easy_keys_in_DL__2d_Lite___2f___20_D
> > iego_Calvenese
> >
> >

Received on Thursday, 10 July 2008 14:27:37 UTC