- From: Aldo Gangemi <a.gangemi@istc.cnr.it>
- Date: Wed, 5 May 2004 19:24:10 +0100
- To: "McBride, Brian" <brian.mcbride@hp.com>
- Cc: "McBride, Brian" <brian.mcbride@hp.com>, SWBPD list <public-swbp-wg@w3.org>
At 16:48 +0100 5-05-2004, McBride, Brian wrote: > > >> OK, my position is that > >Is that representative of consensus in the community? Considering the acceptance rate of our papers and the results of our projects, it seems so :). BTW, considering synsets as concepts is quite common and accepted ... >- provided that we want to transform a >> wordnet into a formal ontology - the semantic interpretation of >> "synset" is that of an equivalence class of words/terms according to >> a common intended meaning. Since "having a unique intended meaning" >> is also applicable to classes, the *default* mapping of synsets is to >> owl:Class. > >Hmm. That logic seems vulnerable to the observation that properties too >have a unique intended meaning, as do instances. Yes, and as I suggested, from a purely logical viewpoint, there is no obligation to use classes or relations. There are best practices from formal ontology engineering though. And at least one of economicity: when using Semantic Web ontology languages, it is preferable to manage a class than a property. >So what are the instances of the class you have in mind? The term 'synset' >suggests it is intended to denote a set of synonyms. Are you suggesting >that the instances of the class are words, with the word 'dog' being of type >Synset(Dog). Or are they dogs? > That's one central issue in SW. Formal ontology best practices tend to distinguish between classes of objects and classes of words. Synsets have a twofold nature: they can be considered *equivalence* classes of words, but based on a *common* intended meaning, and the intended meaning is not about words, but about something else (concepts, things, or even words in the case of metalanguage). Then you can decide to use synsets to manage words. But in this case you get an ontology of word meanings, which approximates the way most ontologies are getting used. A big difference between an ontology extracted from a wordnet and one that isn't, is that the first encodes only the concepts that are lexicalised, i.e. that are in words. > > On the other hand, not only classes have a unique intended meaning, > >You were ahead of me :) > >> but also individuals, and as a matter of fact, many synsets refer to >> individuals like "Italy" or "Cicero". That's why "synset" hasn't a >> precise mapping to formal ontologies. > >Hmm, it seems to a rather ignorant me that equating the obvious synset >containing 'dog' to the class of dogs is a bit dubious in theoretical terms. >Equating it to the class of synonyms of (one sense of) 'dog' seems like a >more accurate representation of the WordNet. And that would be a precise >mapping to a formal ontology, just a different kind of mapping, right? >Later one could consider the relation between the synset containing 'dog' >and the class of dogs. But that is not a sameClassAs relationship. As I've said above, I agree on the possibility of using synsets to create an ontology of the lexicon, but still you should wonder why those words are in the same class, i.e. what is the "identity criterion" you are applying for that class. And - formally - that criterion is the set of properties by which you can define the class of dogs. Again, I agree that a class of synonym words is different from a class of objects or any other type, and that there is a relation between the two. But the benefits of having a comprehensive lexical-oriented ontology of entities is greater than having just an ontology of synonyms (that is there already), then it deserves the additional effort. <snip> > >> Some of them >> (specially some verbs) can be considered as potential >> owl:ObjectProperty, and obviously have hypernyms or hyponyms. >> But I do not encourage this kind of investigation, since for each >> owl:ObjectProperty you can get an owl:Class that reifies it, and this >> is what natural languages do often. E.g., is "GIVE" (as a > > verb-synset) more mappable to an owl:Class (being an action), an >> owl:ObjectProperty (someone gives something), or some OWL-DL >> construct that implements an n-ary relation (someone gives something >> to someone else in a certain way, etc., see Natasha's draft)? We have >> good motivations for each of those interpretations. > >I shall look forward to the draft. Is it a development of Guus' proposal? >However, what you have said so far is causing me to lean towards the view >that trying to treat wordnet as an ontology is a bit iffy and encoding >WordNet in an SW language might be more appropriate. It would be good to >have practical criteria such as concrete use cases or effects on software to >guide a decision. Guus' proposal is a refined datamodel for encoding the original structure of WordNet. I am trying to make a collection of the issues concerning WordNet2SW that have been raised so far, possible actions and solutions, etc. Practical use cases? I have some, Jeremy provided others, anyone else is encouraged to submit > > >> My suggestion is to map any WordNet synset to either an owl:Class or >> to an owl:Individual. > >Is that a fixed mapping, or based on some criteria? no fixed criterion, just heuristics <snip> > >> >> hmm, I stop here, since what I am saying will be part of the report >> to be delivered next week. >> BTW, there seems to be here a nice overlap between [WNAT] (and >> [PORT]), and [OEP], because the interpretations I have given about >> wordnets and thesauri can be considered preliminary sketches for >> ontology "reengineering" patterns, which can be a subclass of >> ontology desing patterns. > >Hmm, we seem to have a tradeoff between just getting a common encoding of >WordNet and tackling some more 'interesting' issues of ontology >reengineering. Do I recall correctly that phase 1 is just about the >encoding? Yes Ciao Aldo -- *;*;*;*;*;*;*;*;*;*;*;*;*;*;*;*;*;*;*;*;*;* Aldo Gangemi Research Scientist Laboratory for Applied Ontology, ISTC-CNR Institute of Cognitive Sciences and Technologies (Laboratorio di Ontologia Applicata, Istituto di Scienze e Tecnologie della Cognizione, Consiglio Nazionale delle Ricerche) Viale Marx 15, 00137 Roma Italy +3906.86090249 +3906.824737 (fax) mailto://a.gangemi@istc.cnr.it mailto://gangemi@acm.org http://www.loa-cnr.it
Received on Wednesday, 5 May 2004 13:24:22 UTC