Re: Owl 1.1 DL and OWL 1.1 Full

From: Jim Hendler <hendler@cs.rpi.edu>
Subject: Owl 1.1 DL and OWL 1.1 Full
Date: Wed, 7 Nov 2007 09:01:52 -0500

> I was thinking about whether this should be an actual issue or not,  
> but decided to start it here - Depending how I interpret the current  
> documents I can see this being anything from a null issue to a real  
> re-chartering need, so I thought I better bring it up...
> 
> History:
>   if you look at the OWL Requirements and Use Cases [1], you will see  
> that "Effective Decision Procedure" (define as decidability) is an  
> Objective of the language, not a requirement.  We may all have  
> somewhat different memories of how we got to that, but basically,  
> there was not consensus in the group at that point in time that every  
> feature we would want in a Web Ontology Language would be decidable.
>   As it turned out, this came to be important, and the case that  
> really brought it to the fore was the issue of inverseFunctional  
> datatypes.  To make a long story short, the discussions around this  
> issue almost scuttled the Working Group (I was on the phone with the  
> W3C director several times to convince him to let us try to find a  
> way around the impasse), and it was a compromise (brokered by Frank  
> van Harmelen) which led to us having a restricted subset of the  
> language, later called OWL DL, and an unrestricted one, which came to  
> be called OWL Full.
>   This distinction is not artificial - the issues that couldn't be  
> resolved at that time, and some still can't, cause the split between  
> the two branches of OWL, and there are clearly those on the AC,  
> including myself (since i'm not chair and can have an opinion this  
> time), who need and care about OWL Full.

Hmm.  I'm not sure that this is the best way to characterize the
distinction.  There was, of course, considerable discussion on whether
there had to be effective reasoners for the entire language or not.  At
the WebOnt F2F in January 2002 there was a vote on a number of proposed
requirements for the language, with a majority in favour of an effective
decision procedure for the entire language but a considerable minority
opposed to the this requirement.  The vote was recorded in the WebOnt
mailing list in message
http://lists.w3.org/Archives/Public/www-webont-wg/2002Jan/0089.html

In early 2002 the results of this vote were turned into the UC&R
document 
http://www.w3.org/TR/2002/WD-webont-req-20020307/
Many of the potential requirements that did not have unanimous support
ended up as objectives in this document.  The section of the document
related to the "effective decision procedure" objective underwent
considerable change during 2002 after the first version of the UC&R
document was published.

Also during 2002 there was the decision to have several versions of OWL,
including OWL DL, which would end up having an effective decision
procedure, and OWL Full, which has undecidable inference.

At the very (!) beginning of 2003 there was a review of the requirements
and the possibility of moving this objective to a requirement for
subsets of the language with effective decision procedures, i.e., OWL
Lite and OWL DL.  I can't find much discussion of this possible change.


I do not belive that inverse functional datatype properties played any
role at all related to this objective, or any significant role in the
split between OWL DL and OWL Full.  There was considerable debate
related to inverse functional datatype properties, but I believe that
the objections to having them in OWL DL was more of a matter of them not
being part of any DL reasoner at the time and the DL reasoner
implementors not being willing to implement this addtion.  

> Issue:
> OK, now to the present - In the OWL 1.1 Web Ontology Language  
> Submission [2], the background makes it very clear that OWL 1.1 arose  
> because "an extension of OWL-DL was proposed."  And, in fact, it  
> turns out that the documents for OWL 1.1 do indeed include an  
> extension to OWL DL, but it is unclear, and a real issue, as to how  
> they treat OWL full.  For example, in the BNF for OWL 1.1, the  
> following is included:
> 
> inverseFunctionalObjectProperty := 'InverseFunctionalObjectProperty'  
> '(' { annotation } objectPropertyExpression ')
> 
> but I can find no similar construct for  
> inverseFunctionalDatatypeProperty - so in essence the OWL Full  
> construct has been ignored completely.

Not so, except, of course, that the OWL 1.1 documents don't treat
with OWL Full much, or even at all.

In OWL Full there is no distinction between object and datatype
properties - all properties are object properties, as stated in Section
5.3 of OWL S&AS (as IOOP = PI)
http://www.w3.org/TR/owl-semantics/rdfs.html#5.3
The status of object and datatype properties in OWL Full (and the
related status of OWL classes and RDF classes) was discussed in the
WebOnt.  There were even official comments on these relationships and
responses and test cases.  The response in  
http://lists.w3.org/Archives/Public/www-webont-wg/2003Sep/0258.html
contains a good characterization of the situation for owl:Class and
rdfs:Class, but the property situation is entirely analogous.

It thus seems to me that the OWL 1.1 proposal is perfectly adequate in
this area.  The fixes needed to the relevant parts of the OWL 1.0
documents in this area are either empty or trivial.

> This could be simple to deal with - these omissions can be easily  
> fixed when a section on OWL 1.1 Full is added to the document
> This could be a charter issue - since it very clearly contradicts the  
> charter [3] statement that
> 
> "All new features should have a clear syntax, and a clear semantics  
> both in terms of OWL DL and OWL Full. The existing compatibility  
> between OWL DL and OWL Full should be preserved, and should be  
> extended to new features wherever possible."

My view is that nothing needs to be done here.  

> Impact:
> So, I'm hoping this will be explained as my misunderstanding of the  
> new documents (which I admit I'm still having trouble working out the  
> details of) and not a real issue.  If I'm right, then I am afraid  
> I'll need to oppose publication of the OWL 1.1 documents until  
> something about this is added somewhere, since I think it would be a  
> mistake for the WG to publish working drafts that are in violation of  
> our charter
> 
>   -JH

> p.s. Please note I only discussed inversefunctionaldatatype, but the  
> same goes for all the differentiators of Full vs. DL - and also for  
> new features, cf. my email about Issue 8 which falls under extending  
> OWL Full in the new feature (i.e. allowing a property chain to end in  
> a datatype property should clearly be allowed in OWL 1.1 Full)

I believe that my analysis is valid in this area as well.

> p.p.s. Lest anyone mistake me - I am not claiming this is currently a  
> charter violation, that would be a big step I'm not ready to take,  
> rather I'm asking for clarity on this and on whether it is something  
> the WG needs to take into consideration.
> 
> [1] http://www.w3.org/TR/webont-req/
> [2] http://www.webont.org/owl/1.1/
> [3] http://www.w3.org/2007/06/OWLCharter.html  (may be member  
> readable only, I forget if charters are public)

Peter F. Patel-Schneider
Bell Labs Research

Received on Friday, 9 November 2007 10:29:50 UTC