- From: Aldo Gangemi <aldo.gangemi@cnr.it>
- Date: Wed, 8 May 2013 23:56:52 -0300
- To: Jorge Gracia <jgracia@fi.upm.es>
- Cc: Aldo Gangemi <aldo.gangemi@cnr.it>, John McCrae <jmccrae@cit-ec.uni-bielefeld.de>, Philipp Cimiano <cimiano@cit-ec.uni-bielefeld.de>, "public-ontolex@w3.org" <public-ontolex@w3.org>
I insist that there is a confusion between semantics of data, and semantics of a domain.Representing words, concepts, or oranges as individuals creates a domain made of words, concepts, and oranges. *If* the ontology-lexicon interface requires to distinguish between words, concepts/meanings and references, we need to satisfy such a requirement. OWL, RDFS or whatever formally interpreted language certainly provides a clear data semantics, but the domain needs to correspond to it in order to get a intensional/extensional coupling. This is Tarskian semantics: "The snow is white" only if the snow is white. When I create a thesaurus, the semantics of the domain is addressed as a pure (and informal) intensional one. Porting it to OWL does not make it formal or extensional. Also, when a SKOS model is punned with concepts representing classes of objects, this is made by accepting the risks of mis-correspondence that derive from that assumption, and anyway they are *no more* thesaurus concepts, but just regular ontology classes that have been reverse-engineered from a thesaurus. The SKOS namespace there is nothing special. Anyway, the point about semiotics is that words, meanings, and objects/collections in the world are different aspects of what we are representing. A thesaurus concept does not become a rabbit more than a potato can become a duck. Of course, we can agree to disagree: please raise your hand if you don't care about the requirement I mentioned at the beginning, and the all discussion will startup again from scratch :) Ciao Aldo On May 8, 2013, at 9:01:41 PM , Jorge Gracia <jgracia@fi.upm.es> wrote: > Dear all, > > I subscribe Philipp's arguments completely. SKOS data model is an OWL > Full ontology and their concepts can be owl instances and/or classes > depending on the circumstances and, as such, could be subject of a > lexical realisation in a lemon/ontolex lexicon. I do not see why not, > or why they should be treated differently. > >> [John:] Informal representations are not really the focus of the group ... > > I wouldn't say that SKOS models are "informal", rather they can be > considered "semi-formal". And, is the dbpedia dataset a "formal" > knowledge representation even if it is written in rdf/owl? I am not so > sure ;-) > > Regards, > Jorge > > > 2013/5/8 John McCrae <jmccrae@cit-ec.uni-bielefeld.de>: >> Hi, >> >> The point is not the SKOS doesn't have an RDF-OWL semantics, of course it >> does! Actually, so does lemon, thus by that reasoning every property in the >> entire lemon model could be replaced with lemon:reference and everything >> would be much less confusing![/sarcasm] >> >> In fact we care about whether the intension/extension of the SKOS concept is >> the same as inferred by the OWL reasoner. In my cat example it was not, I >> intended to express the concept of a cat but was left inferring only the >> genus to which cats belong. >> >> This matters a lot to us, as we want to be able to provide a mapping to and >> from a lexical expression and its formal representation. SKOS is not formal: >> >>> To understand this distinction, consider that the "knowledge" made >>> explicit in a formal ontology is expressed as sets of axioms and facts. A >>> thesaurus or classification scheme is of a completely different nature, and >>> does not assert any axioms or facts. -- >>> http://www.w3.org/TR/2009/REC-skos-reference-20090818/ >> >> >> The SKOS model may be defined by an OWL ontology, but a model expressed >> using SKOS is not an ontology! It is of a completely different nature, a >> informal representation of meaning. >> >> Informal representations are not really the focus of the group other than as >> a bridge to getting to formal representation (i.e., via WordNet synsets). >> More practically, this is because non-formal schemes lack fundamental >> elements (like valence which is very important here >> http://www.w3.org/community/ontolex/wiki/Specification_of_Requirements/Valence_and_Ontological_Mapping).* >> >> To illustrate this consider the case of a word sense disambiguation >> algorithm based on OWL reasoning, in model 2 it would work as follows >> >> Normalize token and identify lexical entry >> For each sense >> >> If there is a ontolex:reference yield its object >> For each ontolex:means (following all skos:exactMatchs thereafter) yield the >> object of ontolex:conceptualizes of this object >> >> Apply OWL reasoning to select best ontology entity >> >> In model 1, it becomes more ambiguous: not every ontolex:reference yields an >> ontology entity and more importantly it is not clear which do. >> >> While I don't think that the difference is humongously important, I would >> say that the OntoLex group should focus on how we link ontologies to lexica, >> if we link ontologies to lexica via KOS/terminologies that may be a >> necessary evil (by which I of course mean a very important and useful use >> case). Our goal is not to be KOSLex or TermLex: but if we can do it neatly >> as a by-product, that's good. As for confusing people, we will only confuses >> people who do not understand/care about the difference between >> formal/informal semantic models, therefore I view using two properties as a >> zero-less strategy, people who care about the difference will get it right >> (because they care and from my experience there are many who do) and people >> who don't care will choose at random but it doesn't matter as their model is >> a mish-mash of formal and informal declarations anyway as they don't care >> (which again many "SKOS ontologies" are)... and if we are very lucky some >> people, who don't care very much, will care just enough to read the >> documentation! >> >> Regards, >> John >> >> * Caveat: SKOS is not the best example as it does provide some formal >> description, e.g., transitiveBroader along side non-formal description, >> e.g., broader >> >> >> >> >> On Wed, May 8, 2013 at 10:39 PM, Aldo Gangemi <aldo.gangemi@cnr.it> wrote: >>> >>> Uhm, you're talking of two different things here. >>> Indeed SKOS in RDF-OWL has a semantics (intensional and extensional), but >>> this is true also for WordNet synsets, senses and words, as well as for any >>> other datum encoded in RDF-OWL. For that reason should we then use only >>> ontolex:reference for everything? >>> And I also see that subtle distinctions can be hard to digest to >>> non-experts. But no distinctions at all is an even worse problem. With the >>> design rationale that any OWL individual can be used as a reference *in >>> general* (in particular cases is of course perfectly fine to treat them all >>> as references), nothing can justify why SKOS concepts are ok and others not. >>> I see two escapes: >>> >>> 1) we maintain the suggestion to use skos:exactMatch for mapping to SKOS, >>> explaining why and how, as it's common in design patterns for software, >>> ontologies, and data (people like being explained useful things sometimes >>> ;)) >>> 2) we collapse the properties, but we create an OWL axiom that states e.g. >>> the following general class axiom: >>> (ontolex:reference some skos:Concept) owl:equivalentTo (skos:exactMatch >>> some skos:Concept) >>> >>> Aldo >>> >>> On May 8, 2013, at 4:52:54 PM , Philipp Cimiano >>> <cimiano@cit-ec.uni-bielefeld.de> wrote: >>> >>> John, >>> >>> of course skos:Concepts have a formal interpretation and thus an extension >>> much as owl:Classes. >>> >>> So I am not assuming: >>> >>> :cat rdf:type skos:Concept ⊨ :cat rdf:type owl:Thing >>> >>> But a SKOS document is an RDF document and thus has an RDF Interpretation >>> which also assign denotations to the URIs in the document. >>> >>> Further, of course SKOS documents has a semantics, which is defined in >>> OWL. >>> >>> For example skos:transitiverBroader is defined as a OWL transitive >>> property. >>> >>> And btw. skos:Concept is defined as being an owl:Class, i.e. skos:Concept >>> rdf:type owl:Class (see axiom S1 in >>> http://www.w3.org/TR/2009/REC-skos-reference-20090818/) >>> >>> So SKOS is as axiomatized as any other OWL ontology ;-) >>> >>> 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. This will compromise the usability of the model IMHO >>> >>> Philipp. >>> >>> Am 08.05.13 21:36, schrieb John McCrae: >>> >>> >>> >>> >>> On Wed, May 8, 2013 at 5:56 PM, Philipp Cimiano >>> <cimiano@cit-ec.uni-bielefeld.de> wrote: >>>> >>>> Dear all, >>>> >>>> I am not with Aldo and John here. >>>> >>>> I think introducing two different properties makes our model >>>> unnecessarily complex. >>> >>> >>> Neither model introduces any new properties, the first allows reference to >>> have two domains (skos:Concept and owl:Thing) but means only one >>> (skos:Concept). The second model uses skos:exactMatch for matching within a >>> SKOS model and differentiates reference and means by domain (owl:Thing and >>> skos:Concept respectively). >>> >>>> >>>> We said we use "reference" when the meaning is expressed by an >>>> extensional entity where we defined extensional as "having an extension in >>>> some model of the theory". I agreed to that. >>>> >>>> According to this, a particular skos:Concept (an individual) has as much >>>> as an extensional interpretation as a particular owl:Class, or an >>>> owl:Individual to stay at the same level. >>> >>> Eh? You are saying: "SKOS concepts have an extension in some model of the >>> theory". This seems very odd, SKOS does not have any formal semantics, >>> therefore how can it have models and extensions? The reasoning seems >>> circular, if I assume SKOS concepts have extensions like an OWL entity then >>> SKOS concepts are like OWL entities because they have extensions. >>>> >>>> >>>> Of course, a particular skos:Concept is an individual from an RDF/OWL >>>> perspective and is also interpreted as some individual in the corresponding >>>> domain, much like an owl:Individual. So a model assigns some extensional >>>> interpretation to both skos:Concepts and owl:Individuals. Where is then the >>>> essential difference that prevents us using the same property for both then? >>> >>> Again, you are assuming that as >>> >>> :cat rdf:type skos:Concept ⊨ :cat rdf:type owl:Thing >>> >>> Ergo, :cat has an extension in OWL. But this only true because you applied >>> an OWL reasoner, and more importantly the meaning has changed: the OWL >>> reasoner has only indicated that there are is a (individual) thing (the >>> genus of cats), where as the SKOS concept intended an extension as the class >>> of cats. >>> >>>> >>>> >>>> Surely, skos:Concept are per definition "intensions", but technically >>>> they are extensional entities according to our definition, i.e. >>>> owl:Individual or rdf:Thing. >>>> >>>> Further, it is perfectly fine for a skos:Concept to be an owl:Class (see >>>> http://www.w3.org/TR/2009/REC-skos-reference-20090818/) >>> >>> >>> True, but more importantly a skos:Concept can also be an >>> owl:DatatypeProperty, an owl:ObjectProperty and an owl:Thing. The difference >>> is the underspecification. >>> >>>> >>>> What do we use then? "reference" or "means"? ;-) >>>> >>>> >>>> Treating skos:Concept and owl:Class as different types of meaning seems >>>> too subtle for people who want to use the model in practice as they will >>>> always wonder which is the right property to use. >>> >>> >>> Regards, >>> John >>> >>>> >>>> Philipp. >>>> >>>> Am 08.05.13 13:08, schrieb John McCrae: >>>> >>>> Hi Aldo, >>>> >>>> Names in the previous example are not fixed of course. I also don't like >>>> "means" that much I just haven't got a better alternative yet. (synset is >>>> too WordNet-specific, means/meaning/concept are too broad) >>>> >>>> Regards, >>>> John >>>> >>>> >>>> On Wed, May 8, 2013 at 1:02 PM, Aldo Gangemi <aldo.gangemi@cnr.it> wrote: >>>>> >>>>> Hi, I agree with John, we really seem on the same wave now :), in fact I >>>>> agree with Model 2 being far better. >>>>> Only, should we really use ontolex:means to link senses and synsets? >>>>> It's a bit too broad as a name for a specific relation like that, isn't it? >>>>> >>>>> Aldo >>>>> >>>>> On May 8, 2013, at 6:37:22 AM , John McCrae >>>>> <jmccrae@cit-ec.uni-bielefeld.de> wrote: >>>>> >>>>> Hi Jorge, all, >>>>> >>>>> Thanks for your comment, I agree this is an issue we should discuss. I >>>>> think that it is clearly wrong to continue to treat skos:Concepts as >>>>> ontological elements, they aren't and we shouldn't really confuse them. The >>>>> question of whether we should still use SKOS terminologies as systems of >>>>> reference for the model also seems clear to me (of course we should). >>>>> >>>>> The question then boils down to this essential question: do we use the >>>>> same property to reference both a skos:Concept and an ontology entity? >>>>> >>>>> This leads to two variation on the model: >>>>> >>>>> Model 1. (Same property) >>>>> >>>>> With synsets >>>>> >>>>> :corn --ontolex:sense-> :corn_sense1 --ontolex:means-> >>>>> wordnet:corn_n_xxx --ontolex:conceptualizes-> fao:Corn (a skos:Concept) >>>>> :corn --ontolex:sense-> :corn_sense1 --ontolex:means-> >>>>> wordnet:corn_n_xxx --ontolex:conceptualizes-> dbpedia:Corn (a owl:Class) >>>>> >>>>> Without synsets >>>>> >>>>> :corn --ontolex:sense-> :corn_sense1 --ontolex:reference-> fao:Corn (a >>>>> skos:Concept) >>>>> :corn --ontolex:sense-> :corn_sense1 --ontolex:reference-> dbpedia:Corn >>>>> (a owl:Class) >>>>> >>>>> Model 2. (Different property) >>>>> >>>>> With synsets >>>>> >>>>> :corn --ontolex:sense-> :corn_sense1 --ontolex:means-> >>>>> wordnet:corn_n_xxx --skos:exactMatch-> fao:Corn (a skos:Concept) >>>>> :corn --ontolex:sense-> :corn_sense1 --ontolex:means-> >>>>> wordnet:corn_n_xxx --ontolex:conceptualizes-> dbpedia:Corn (a owl:Class) >>>>> >>>>> Without synsets >>>>> >>>>> :corn --ontolex:sense-> :corn_sense1 --ontolex:means-> fao:Corn (a >>>>> skos:Concept) >>>>> :corn --ontolex:sense-> :corn_sense1 --ontolex:reference-> dbpedia:Corn >>>>> (a owl:Class) >>>>> >>>>> With further linking valid of >>>>> >>>>> fao:Corn --ontolex:conceptualizes-> dbpedia:Corn >>>>> >>>>> >>>>> I prefer model two as it makes a clearer distinction between >>>>> terminologies and ontologies, doesn't require linking two SKOS concepts with >>>>> an ontolex property (which we should avoid as it is not our job to fix SKOS) >>>>> and allows us to define a natural property for linking terminologies to >>>>> ontologies. >>>>> >>>>> Regards, >>>>> John >>>>> >>>>> >>>>> >>>>> On Fri, May 3, 2013 at 2:58 PM, Jorge Gracia <jgracia@fi.upm.es> wrote: >>>>>> >>>>>> Dear Philipp, all >>>>>> >>>>>> I am not able to join the telco today, sorry. But let me to formulate >>>>>> a quick question about John's model >>>>>> http://www.w3.org/community/ontolex/wiki/File:John-modelling.png); >>>>>> maybe you can treat it today. >>>>>> Following the previous discussions I can understand the inclusion of >>>>>> the new class "Synset / Concept". My doubt is: despite the fact that >>>>>> skos concepts could be represented with this new class, can we >>>>>> alternatively continuing treating skos concepts (of external skos >>>>>> ontologies) as "ontology entities"? (as in the IFLA example presented >>>>>> last week). For me this option is very natural, fully compliant with >>>>>> R3 "semantics by reference" and we shouldn't lose it. >>>>>> >>>>>> Best regards, >>>>>> Jorge >>>>>> >>>>>> >>>>>> 2013/5/2 Philipp Cimiano <cimiano@cit-ec.uni-bielefeld.de>: >>>>>>> Dear all, >>>>>>> >>>>>>> this is a gentle reminder that we will have our regular ontolex >>>>>>> telco >>>>>>> tomorrow. >>>>>>> >>>>>>> I intend to discuss the model proposed by John on the basis of the >>>>>>> contributions of all of you. >>>>>>> I would like to see if there is a chance that we agree on this model >>>>>>> as a >>>>>>> building block for the further work. >>>>>>> >>>>>>> Here is a link to the conference metadata including access details: >>>>>>> >>>>>>> >>>>>>> http://www.w3.org/community/ontolex/wiki/Teleconference,_2013.03.05,_15-16_pm_CET >>>>>>> >>>>>>> Best regards, >>>>>>> >>>>>>> Philipp. >>>>>>> >>>>>>> -- >>>>>>> Prof. Dr. Philipp Cimiano >>>>>>> Semantic Computing Group >>>>>>> Excellence Cluster - Cognitive Interaction Technology (CITEC) >>>>>>> University of Bielefeld >>>>>>> >>>>>>> Phone: +49 521 106 12249 >>>>>>> Fax: +49 521 106 12412 >>>>>>> Mail: cimiano@cit-ec.uni-bielefeld.de >>>>>>> >>>>>>> Room H-127 >>>>>>> Morgenbreede 39 >>>>>>> 33615 Bielefeld >>>>>>> >>>>>>> >>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> Jorge Gracia, PhD >>>>>> Ontology Engineering Group >>>>>> Artificial Intelligence Department >>>>>> Universidad Politécnica de Madrid >>>>>> http://delicias.dia.fi.upm.es/~jgracia/ >>>>>> >>>>> >>>>> >>>> >>>> >>>> >>>> -- >>>> Prof. Dr. Philipp Cimiano >>>> Semantic Computing Group >>>> Excellence Cluster - Cognitive Interaction Technology (CITEC) >>>> University of Bielefeld >>>> >>>> Phone: +49 521 106 12249 >>>> Fax: +49 521 106 12412 >>>> Mail: cimiano@cit-ec.uni-bielefeld.de >>>> >>>> Room H-127 >>>> Morgenbreede 39 >>>> 33615 Bielefeld >>> >>> >>> >>> >>> -- >>> Prof. Dr. Philipp Cimiano >>> Semantic Computing Group >>> Excellence Cluster - Cognitive Interaction Technology (CITEC) >>> University of Bielefeld >>> >>> Phone: +49 521 106 12249 >>> Fax: +49 521 106 12412 >>> Mail: cimiano@cit-ec.uni-bielefeld.de >>> >>> Room H-127 >>> Morgenbreede 39 >>> 33615 Bielefeld >>> >>> >> > > > > -- > Jorge Gracia, PhD > Ontology Engineering Group > Artificial Intelligence Department > Universidad Politécnica de Madrid > http://delicias.dia.fi.upm.es/~jgracia/
Received on Thursday, 9 May 2013 02:57:25 UTC