Re: doubt about "Synset / Concept" class

Dear all, let me recap this discussion in terms of requirements we can endorse or not.

Previously, I asked if anyone endorses or not the requirement (0):

 (0) distinguishing expressions, meanings, and references as disjoint *aspects* (not things) of the ontology-lexicon interface. I call it the "semiotic stance" for the interface between linguistic and formal semantics.

Probably John supports the requirement based on his recent emails.

Alessandro has answered that he does not endorses it, and attributes to Philipp the same conviction based on his assertion:

 "So we should not treat SKOS concept as being really different from OWL concepts at the lexicon level. Such distinctions are very subtle and people will not grasp this difference, having doubts on which property to use in a particular case."

I think Phiipp's assertion is not about the general requirement I asked for, but on another requirement (3) that I formulate below.

The position of the others is less clear or they did not comment.

Coming to the specific discussion on SKOS, it seems clear that we have the following requirements on the table:

 1) being able to map lexical concepts to SKOS concepts (everyone supports it I guess :))

 2) being able to treat SKOS as regular ontologies (I guess many support it, and me too!)

 3) being able to distinguish between OWL encodings of purely intensional concepts, and OWL encodings of concepts bearing a clear extensional intepretation in the domain that they help conceptualizing (I and probably John support this).

Notice that (3) is a requirement not limited to SKOS, but it holds also for lexica, NLP results, folksonomies, etc. 

To remind about the difference: an owl:Class onto:Dog in an ontology about dogs is supposed to have an intensional interpretation (the concept), as well as an extensional interpretation including all dogs (within an open world in OWL). 

A skos:Concept myskos:Dog is only supposed to have an intensional intepretation, and in fact it is an individual in the domain of concepts, not of dogs.

A lexical concept is similar to a SKOS concept, then it'd be straightforward to map to it as a concept/meaning, whatever property we are using, although skos:exactMatch seems pretty good for that. 

Now, two new possible requirements, orthogonal to (3):

 (4) being able to jointly reason on concepts and classes
 (5) being able to reason on concepts *as* classes

(4) is granted by any relation between concepts (skos or lexical), and classes, modulo the limits in interpretation when individuals and classes appear in object property triples.

About (5), when I pun the constant myskos:Dog as an owl:Class, I'm simply extending its intepretation to the extensional, domain-oriented side, and that class has nothing in common (except the name) with the original SKOS concept. Similarly for lexical concepts, and in fact, none here is suggesting to create (by default) an owl:equivalentClass mapping between a lexical concept and an OWL class from an ontology. 

(5) should be discussed in our proposed ontology as a measure that can be taken by those willing to extend the semantics of concepts/meaning for their own reasons. Punning would be the right technical means to do that.

I hope this helps to gets us a bit more focused.
Aldo

On May 10, 2013, at 8:15:07 AM , Jorge Gracia <jgracia@fi.upm.es> wrote:

> Hi John,
> 
>> Firstly, we should bear in mind that SKOS is not an ontology:
> 
> Well, ontologies are resources representing the conceptual model
> underlying a certain domain, describing it in a declarative fashion
> cleaningly separated from procedural aspects [1]. In that sense, and
> also in Grubber's traditional definition [2], SKOS models can be
> considered ontologies. Although, of course, they are not "formal"
> ontologies (the spectrum of formality is wide, as in Lassila's
> classification [3]).
> 
>> As such, we have two ways to go as a group. We could be in the very
>> prescriptivist camp of saying "SKOS models are not ontologies, we should
>> explicitly tell people they have to link to SKOS models differently".
>> Alternatively, we could be in the permissive camp "Many people use SKOS as
>> ontologies, we should accommodate this... (and SKOS is close enough to
>> ontologies that the difference does not really matter)". This is therefore
>> just a question of documentation, and perhaps the best solution is to say
>> nothing at all (the prescriptivists aren't upset and the permissivists are
>> not disallowed).
> 
> Yes, I agree that is the best option (avoid commitments to any view),
> although we still have to decide which domain/ranges will have our
> ontology-lexicon mapping properties (so some commitments made).
> 
>> Stated equivalence between symbols, e.g., (onto:SamuelClemens owl:sameAs onto:MarkTwain)
>> Inferred equivalence between symbols, e.g., (onto:Yeti ≡ ⊥ ∏ GreatestNaturalNumber ≡ ⊥ ⊨Yeti ≡ GreatestNaturalNumber)
> 
> I am not sure if you can entail that two concepts are the same if they
> share the same (empty) set of instances.
> 
> Regards,
> Jorge
> 
> ----
> 
> 
> [1] Philipp Cimiano. Ontology learning and population from text:
> Algorithms, evaluation and applications. Springer, October 2006
> [2] Thomas R. Gruber. Toward principles for the design of ontologies
> used for knowledge sharing. International Journal Human Computer
> Studies, vol. 43, pages 907-928, November 1995.
> [3] Ora Lassila & Deborah L. McGuinness. The Role of Frame-Based
> Representation on the Semantic Web. Technical report, Knowledge
> Systems Laboratory Stanford University, KSL-01-02, 2001
>

Received on Friday, 10 May 2013 12:04:49 UTC