Re: Publications about OWL (1 or 2) Full

On 19/05/11 18:58, Antoine Zimmermann wrote:
> First, thanks to you Michael and Markus for your replies.
> Now, Michael,
>>> Fortunately, OWL 2 now allows a useful form of simple meta-modelling
>>> now,
>>> so that you can indeed have meta classes and use classes as subjects and
>>> objects of properties.
>> The logical inferences that OWL 2 DL tools draw from this are limited,
>> but
>>> may still be more than what any particular OWL 2 Full reasoner would
>>> give
>>> you (depends on the OWL 2 Full reasoner you have -- I am not aware of
>>> much
>>> implementation work there beyond OWL 2 RL).
>> Hmm, I know there is some limited punning, but these are two different
>> things, not one thing appearing in two different places. The inference is
>> very limited.
> What Markus says here I guess is that, in spite of the limitations of
> the punning mechanism, a full-fledged OWL 2 DL reasoners will likely
> infer more things than *currently existing* incomplete OWL Full reasoners.

Right. We know that there cannot be a tool that computes all 
consequences of OWL with "proper" meta modelling, and we also know that 
some forms of meta modelling can even lead to intricate inconsistencies 
that make the whole ontology language paradoxical (PF Patel-Schneider's 
paper "Building the Semantic Web Tower from RDF Straw" alludes to this 
issue). So it seems that a tool that obtains all consequences of plain 
OWL constructs, and that can still handle some meta modelling is not 
such a bad choice, even if it is called "OWL DL reasoner" ;-)

>> I don't think there is a way to nicely handle the species example where
>> Species is a class with instance Eagle with instances being individual
>> eagles.
> No problem:
> :Species a owl:Class .
> :Eagle a :Species, a owl:Class ;
> rdfs:subClassOf :Animal .
> :billy a :Eagle .
> This is valid OWL 2 DL.
> Then, with a SPARQL 1.1 query with OWL 2 DL entailment regime, I can get
> the pairs <species,individualmemberofthespecies>:
> SELECT ?species, ?member WHERE {
> ?species a :Species .
> ?member a ?species .
> }

Yes, this is allowed.

>> I also do not think there is a robust solution to the classes as values
>> problem.
> What do you mean by "classes as values problem"?
>>> An insightful discussion of meta modelling semantics -- the one of
>>> OWL 2 DL
>>> (punning) and a stronger one -- is found in the paper:
>>> Boris Motik. On the Properties of Metamodeling in OWL. Journal of
>>> Logic and
>>> Computation, 17(4):617–637, 2007.
>> Thanks, I just had a look. It is intersting, and geared more for the
>> theorist than the practitioner. Do you know of a more practice-focused
>> paper that gives examples of what you can and cannot do with OWL2
>> metamodelling, compared to OWL-Full?

Indeed, this paper is more on the logical side of the discussion, though 
I still found it quite accessible. Especially, it has some examples of 
consequences that one looses under the weak meta modelling of OWL 2.

I am not aware of a treatment of this issue that is using OWL or RDF 
terminology. This may not make it easier to understand, since the issues 
of metamodelling are often complicated by nature -- the straw tower 
paper mentioned above uses the RDF data model but still requires some 
thought to understand the key issues raised there.

>>> A big advantage of OWL 2 DL in this respect is that it makes it legal to
>>> state such meta-knowledge without violating any constraints of the
>>> language.
>>> The OWL Full semantics may still formally lead to more consequences,
>>> but in
>>> practice what matters is how many of the total consequence any tool will
>>> actually give. So the DL approach could be a good compromise
>>> (especially to
>>> "make meaning clear" beyond purely logical/formal aspects).
>> I'm not sure what you mean by "make meaning clear" as a good DL
>> compromise.
>> The example from that paper is the need to represent Eagle as an instance
>> of Species so you can e.g. say it is on the engangered list. DL forces
>> you
>> to represent Eagle as an as an individual that can not ever have any
>> instances. But this is patently untrue -- to that extent, it obfusticates
>> meaning. If OWL2 metamodellign lets me do this, I'll be surprised and
>> delighted.
> Punning means that you can use the URI of an individual in place of the
> URI of a class. Therefore, :Eagle, as a class, can have instances (like
> :billy above) and as an individual it can belong to a class (like
> :Species). However, :Eagle-the-individual is different from
> :Eagle-the-class, although they share the same identifier.

Exactly. This is of course a cheap form of meta modelling, but it seems 
that it goes a surprisingly long way in practice. Many use cases are 
really about modelling several "layers" of the domain of interest, but 
have only little interaction between these layers. Here is an example 
where one would see the limitation:

Assume you have Eagle and Hawk as classes, and you have an individual 
Tweety who is said to have the species Eagle, and to have the species 
Hawk (as individuals). Assume further that there is a cardinality 
restriction that requires "has species" to be functional. Then 
implicitly we derive that Eagle and Hawk are the same individuals. With 
punning, nothing else happens. With "true" meta modelling, the classes 
Eagle and Hawk would also be inferred to be the same, with all the 
consequences that this could have.

I am not sure if this is a practically relevant limitation.



>>> I think the more important case where ontologies go beyond OWL DL is
>>> due to
>>> the structural constraints related to transitivity and property
>>> chains (e.g.
>>> it is easy to get forbidden cycles in property chain dependencies).
>>> But the
>>> interesting difference to the earlier meta-modelling limitations of
>>> OWL 1 DL
>>> is that in these cases, the semantics of OWL DL is in principle still
>>> meaningful and well-defined in its common first-order logic
>>> framework. It is
>>> simply known that computing consequences of this semantics becomes
>>> undecidable, and thus the decidability-loving DL tools reject the inputs
>>> right away.
>>> But again anybody who would venture to implement OWL Full reasoning
>>> could
>>> also look into "OWL DL reasoning for ontologies violating the structural
>>> restrictions." This task might be easier to solve in practice since one
>>> could probably reuse existing algorithms and tools to solve part of the
>>> problem. It is also part of ongoing research to weaken the structural
>>> restrictions further, so one already knows of complete algorithms
>>> that could
>>> achieve this in some cases that OWL DL excludes.
>>> Also note that "FULL" and "DL" now refer to syntactic languages only.
>>> The
>>> semantic distinction is now made between "direct semantics" and
>>> "RDF-based
>>> semantics". This helps a bit to avoid confusion between syntax and
>>> semantics. So my last remark was about finding ways to evaluate (more
>>> of)
>>> OWL 2 FULL under direct semantics.
>>> Cheers,
>>> Markus
>>>> I have no hard evidence, but I feel certain that there are plenty of
>>>> cases when the penalties of OWL Full are on balance small enough
>>>> compared to the gains of expressive convenience and clarity of OWL
>>>> Full.
>>>> I would love to see someone look into this. I would love it if someone
>>>> tried to create a reasoner that handled OWL Full as efficiently as
>>>> possible.
>>>> Notice how many responses you got to this message in the past few
>>>> weeks?
>>>> That may reflect how much people in the community care about OWL Full!
>>>> Michael
>>>> Michael
>>>> On Sat, Apr 23, 2011 at 1:05 AM, Antoine Zimmermann
>>>> <
>>>> <>> wrote:
>>>> Dear all,
>>>> I'm looking for scientific publications related to OWL Full. I'm
>>>> interested in the following kind of work:
>>>> - reasoning with OWL Full;
>>>> - modelling ontologies in OWL Full;
>>>> - properties of OWL Full, or relationships between OWL Full and
>>>> other formalisms.
>>>> I've found some papers about modelling existing ontologies in OWL
>>>> (for instance, modelling a UML spec or a frame-based ontology in
>>>> OWL) which happen to fall into OWL Full, but nothing about modelling
>>>> OWL Full ontologies by design. I found very little about reasoning
>>>> in OWL Full (with the notable exception of [1], which also relates
>>>> OWL reasoning to OOP).
>>>> But the vast majority of papers mentioning OWL Full present it as
>>>> the language that must be avoided at all cost (usually saying "if we
>>>> do that, we are in OWL Full" implying "if we do that, we're screwed!").
>>>> Thanks in advance for your pointers.
>>>> [1] Seiji Koide and Hideaki Takeda. OWL-Full Reasoning from an
>>>> Object Oriented Perspective. In R. Mizoguchi, Z. Shi, and F.
>>>> Giunchiglia (Eds.): ASWC 2006, LNCS 4185, pp. 263–277, 2006.
>>>> Springer-Verlag Berlin Heidelberg 2006.
>>>> Regards,
>>>> --
>>>> Antoine Zimmermann
>>>> Researcher at:
>>>> Laboratoire d'InfoRmatique en Image et Systèmes d'information
>>>> Database Group
>>>> 7 Avenue Jean Capelle
>>>> 69621 Villeurbanne Cedex
>>>> France
>>>> Tel: +33(0)4 72 43 61 74<tel:%2B33%280%294%2072%2043%2061%2074> -
>>>> Fax: +33(0)4 72 43 87 13<tel:%2B33%280%294%2072%2043%2087%2013>
>>>> Lecturer at:
>>>> Institut National des Sciences Appliquées de Lyon
>>>> 20 Avenue Albert Einstein
>>>> 69621 Villeurbanne Cedex
>>>> France
>>>> --
>>>> Michael Uschold, PhD
>>>> Senior Ontology Consultant, Semantic Arts
>>>> LinkedIn:
>>>> Skype, Twitter: UscholdM
>>> --
>>> Dr. Markus Krötzsch
>>> Oxford University Computing Laboratory
>>> Room 306, Parks Road, Oxford, OX1 3QD, UK
>>> +44 (0)1865 283529

Dr. Markus Krötzsch
Oxford  University  Computing  Laboratory
Room 306, Parks Road, Oxford, OX1 3QD, UK
+44 (0)1865 283529

Received on Thursday, 19 May 2011 19:02:01 UTC