RE: LC Comment: "Hidden" Axioms

Dear all,

This discussion reminds me of a related issue that is more
designer-oriented.

Analogously to labelling OWL 2 axioms as 'in/active', there could also be a
similar label (annotation) indicating whether the axiom is, by the
assessment of its creator, 'firm' or somewhat 'weakened' (in various
senses).
It looks like a notorious situation in ontological engineering that certain
axioms 'essentially hold', but there is some small proportion of cases (or,
some special context) where the axiom does not hold.
For a trivial example:
- a firm axiom could be "Person subclassOf (hasBiologicalParent exactly 2)"
- a weakened axiom could be "Person subclassOf (marriedTo max 1)" -
considering polygamy.

There are currently afaik three solutions to this problem:
- to give up the axiom (to remain completely ontologically safe)
- to refine the ontology so as to explicitly address the exception
- to resort to some (quantitative) uncertainty modelling formalism.

The proposed 'axiom flagging' solution would allow to easily (by checking a
single box in an editor) indicate that there should be 'some caution' made
for the given axiom. It would then be up to the reasoner to either ignore
this flag or take it somehow into account - e.g. by looking at some
additional (more specific) annotation from some annotation space it is
aware of.

I discussed this a bit with Bijan at OWLED EU. My plan was to first collect
some empirical evidence to motivate this feature, but the current
discussion made me mention it already now, just in case.

Regards
Vojtech

----------------------------------------------------------------------
Vojtech Svatek, University of Economics, Prague
Nam.W.Churchilla 4, 13067 Praha 3, CZECH REPUBLIC
phone: +420 224095495, e-mail: svatek@vse.cz
web: http://nb.vse.cz/~svatek



-----public-owl-wg-request@w3.org napsal: -----


Komu: "Bijan Parsia" <bparsia@cs.manchester.ac.uk>
Od: "Solbrig, Harold R." <Solbrig.Harold@mayo.edu>
Odeslal: public-owl-wg-request@w3.org
Datum: 24.01.2009 00:11
Kopie: "W3C OWL Working Group" <public-owl-wg@w3.org>
Předmět: RE: LC Comment: "Hidden" Axioms



-----Original Message-----
From: Bijan Parsia [mailto:bparsia@cs.manchester.ac.uk]
Sent: Friday, January 23, 2009 4:58 PM
To: Solbrig, Harold R.
Cc: W3C OWL Working Group
Subject: Re: LC Comment: "Hidden" Axioms

On 23 Jan 2009, at 22:46, Solbrig, Harold R. wrote:

[snip]

>> Let me mention related requirement that we have encountered.  We
>> have a
>> need to be able to classify an ontology and subsequently transmit
both
>> the asserted and the logical inferences for display and consumption
in
>> secondary resources such as wikis and other tools.  As these tools
may
>> create additional axioms, we need to differentiate the asserted from

>> the
>> inferred - both as important information to the editors and to be
able
>> to remove or ignore these axioms when the modified ontology is
>> subsequently re-classified. While this is a slightly different use
>> case,
>> it still involves the same notion - some sort of tag or property on
an
>> axiom that affects the way that it is interpreted.

> Actually, from a logical point of view the added inferred axioms are
> harmless...they were already "in" the annotation. So a simple
> annotation which indicates that it was inferred would be fine ---
> editors could not display them, or try to verify them, but if some
> tool passed it to a reasoner...the reasoner would just have an easier

> time :)

True only if the editors didn't add (or remove) one or more non-inferred
axioms.  Then, it would be important to strip the inferred axioms before
re-classifying.  Obviously, this could be done pre-reasoner as long as
the inferred tag was available.

> (Of course, if you were trying to *test* the reasoner...putting in the

> entailments from some other reasoner wouldn't be good. So there's
> still some justification there. I just think it's somewhat less
severe.)

Interesting point. The "inferred" tag would allow you to combine the
input and result of a test cases into a single source.  Instead of an
input and result, just a one set of defined and inferred axioms.

Cheers,

Harold Solbrig

Received on Saturday, 24 January 2009 20:30:19 UTC