- From: Frank van Harmelen <Frank.van.Harmelen@cs.vu.nl>
- Date: Wed, 19 Dec 2001 01:19:01 +0100
- To: Mike Dean <mdean@daml.org>
- CC: Guido_Vetere@tivoli.com, Nicola.Guarino@ladseb.pd.cnr.it, www-archive@w3.org
Mike (Cc Guido and Nicola), Below input from Guido Vetere on suggestions for the Web Ontology language, as input for the January meeting of the Working Group. He raises some very interesting issues not raised by others before: - whether other relations then "subclassOf" (such as "partOf") should be included in the language (he argues cogently why not). - He does argue to introduce some other distinctions, though. In particular between "essential" and "non-essential" roles. (I believe this is closely connected to the remarks by Dan Brickley on time-varying roles, in his recent message to the WebOnt group at http://lists.w3.org/Archives/Public/www-webont-wg/2001Dec/0140.html) - the need for better supporting natural-language "names" for the objects in the language than is currently available in RDF. (Guido, Nicola: Mike has taken over my role to collect input for the meeting, which is why I forward your contribution to him). Frank. ---- -------- Original Message -------- Subject: SIG2 Amsterdam meeting - summary Date: Tue, 11 Dec 2001 16:56:54 +0100 From: Guido_Vetere@tivoli.com To: Frank.van.Harmelen@cs.vu.nl CC: Nicola.Guarino@ladseb.pd.cnr.it Hi Frank, here is a brief summary of my remarks on the SIG2 meeting, to be used as an input for the W3C ongoing discussions. 1 - Modeling 'part' relations: formal vs substantial features (whether the ontology language should include the 'part' relation or not). In my opinion, although mereology is a very formal theory, 'part' is not a formal relation itself. In fact, its meaning depends on the 'particulars' it applies to: sortals, material entities in general, collections, space-time locations, and so on. As a formal language, the W3C ontology syntax should avoid the introduction of such non-formal elements. On the other hand, classic relational properties (transitivity, reflexivity, etc) should be available for defining (standard) 'top level' relations, such as 'material-part', 'cause', etc. More in general, I think that a formal ontology language, which is typically grounded on individuals, sets, relations, etc, should also deal with a certain number of 'universals', such as 'identity' or 'dependency'. For instance, it would be interesting to have classes divided in 'essential' and 'non-essential' (types/roles), the former providing their instances with an identity criterion. Such a way, 'person' (type) would be formally different from 'worker' (role) as, for instance, 'Guido Vetere' can stop being a worker, but can't cease to be a person (while keeping his identity). Note that such a distinction would provide a constraint that leads to a clean structuring of the ontology content. For a deep discussion of these topics, please refer to Nicola Guarino's (cc-ed) 'OntoClean' method. 2 - Promoting the 'Sign' as a first-class object. Even if natural language is by far the most important interface to the web, just a little attention is currently paid to the problem of attaching lexical information to ontologies. RDFS gives NL a very poor support, based on the assumption that literal labels must be directly attached to classes. Such a way, providing different lexicalizations for the same conceptual schema (e.g. 'novice dictionary' vs 'skilled dictionary') turns out to be quite hard. My idea is to promote the 'Sign' (i.e. the union of a 'signifier' and a 'signified') as a 'first-class' object. The following example shows the benefits of this approach: The definition: <rdfs:Class rdf:ID="Resource"> <rdfs:label xml:lang="en">Resource</rdfs:label> <rdfs:label xml:lang="it">Risorsa</rdfs:label> </rdfs:Class> could be written as: <rdfs:Class rdf:ID="Resource"/> ... <xxx:Word xxx:cat="Noun" xxx:ID="N_resource" xml:lang ="en">Resource</xxx:Word> <xxx:Word xxx:cat="Noun" xxx:ID="N_resource" xml:lang ="it">Risorsa</xxx:Word> ... <yyy:Sign> <yyy:signified="http://.../ontology#Resource"/> <yyy:signifier="http://.../lexicon/en#N_resource"/> </yyy:Sign> ... <yyy:Sign> <yyy:signified="http://.../ontology#Resource"/> <yyy:signifier="http://.../lexicon/it#N_resource"/> </yyy:Sign> the advantage here is to have the lexicon separated from the ontology. A collection of Signs would provide the semantics to one or more dictionaries, with reference to one or more ontologies. Moreover, many authors in semiotics (semiology) could be investigated for introducing different types of Sign, as well as relationships among Signs. I think that many problems of conceptual modeling are, actually, semiotic problems. I should try to write something on that .. Hope that you find this note useful .. keep me in the loop. Regards, Guido PS. my mail address gvetere@it.ibm.com is temporarily unavailable, pls use gvetere@tivoli.com instead
Received on Tuesday, 18 December 2001 19:19:41 UTC