Re: Handling multiple rdfs:ranges

Some more inline comments again.


On 23/02/2016 12:55, Ross Horne wrote:
> Hi Fabian, Reto and Graham,
>
> That is good news that most ontologies use owl:UnionOf and somewhere
> Bioportal does the same. It may be somewhere else in the stack that is
> causing queries such as the following:
>
> PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
> PREFIX bpo: <http://www.semanticweb.org/ontologies/2010/10/BPO.owl#>
>
> SELECT ?x
> WHERE {
> bpo:has_event rdfs:domain ?x .
> }
>
> to return multiple types, where it should return one owl:unionOf
> collection, as Graham describes.
>
> Reto's suggestion of using schema:rangeIncludes matches the *some*
> semantics, informally at least. Do you know whether there is an
> established model theory for the semantics of this predicate; and,
> perhaps more importantly, whether there is significant adoption?
>
> Regarding monotonicity, if we take AZ's nice example again. Assume that
> we design a dataset using ex:myProperty rdfs:domain ex:Person,
> containing Graham (for comic effect only, no offence) as the subject of
> a triple with property ex:myProperty. In an open world, someone can come
> along and extend the ontology with ex:myProperty rdfs:domain ex:Woman;
> and now.. Graham is inferred to be a woman. The hypothetical weaker
> semantic for multiple domain/range wouldn't make the same mistake.

Let's just avoid the potential for unintended offense and call the 
person Andrea. How is it a mistake to deduce that some entity of which 
you only have an identifier is a woman? BTW, there may be actual persons 
whose name is Graham and who are female.

In any case, if you have knowledge that Andrea (or Graham) is a man, 
this can be made explicit, or perhaps inferred by other means. The 
disjointness of men and women (though potentially controversial in some 
extraordinary cases) could be made explicit, and the mistake highlighted 
automatically.

However, changing the semantics of rdfs:range, or using a different 
property such as schema:rangeIncludes, would not be very helpful. With 
such semantics, we would know that Andrea (or Graham) is a person, or 
maybe something else. Everything is either a person or something else. 
No knowledge is added with such statement.

Schema.org somehow assumes /total knowledge/ with /local data/ in its 
description of property "rangeIncludes". It assumes that we know 
everything about the property described, especially all the types that 
can be used, and it assumes that the data found where the property is 
used is all that exists about it on the Web. Such assumptions is a 
powerful heuristic that works very well in many practical cases, but it 
is not very Web-compliant, in my opinion.


--AZ

>
> Is this the kind of monotonicity problem you mean, or was there a
> different monotonicity problem in fact in favour of the stronger semantics?
>
> Best regards,
>
> Ross
>
>
>
> On 23 February 2016 at 18:49, Fabian Cretton <Fabian.Cretton@hevs.ch
> <mailto:Fabian.Cretton@hevs.ch>> wrote:
>
>     As Graham explains, I think many ontologies now use the owl:unionOf
>     construct to handle mutliple ranges.
>     And so do tools as Protégé.
>     See for instance the schema.org <http://schema.org> ontology[1]
>     where mutliple domain/range are often used.
>     Here is an example:
>     schema:about a rdf:Property;
>          rdfs:label "About"@en <mailto:%22About%22@en>;
>          rdfs:comment "The subject matter of the content."@en
>     <mailto:%22The+subject+matter+of+the+content.%22@en>;
>          rdfs:domain [ a owl:Class; owl:unionOf
>     (schema:CommunicateAction schema:CreativeWork) ];
>          rdfs:range schema:Thing;
>     The Bioportal  ontology certainly does so as well.
>     Fabian
>
>     [1] http://schema.rdfs.org/
>      >>> Graham Klyne <gk@ninebynine.org <mailto:gk@ninebynine.org>>
>     23.02.2016 11:35 >>>
>
>     On 23/02/2016 09:24, Ross Horne wrote:
>      > My follow up question is: whether anyone knows whether the more
>      > accommodating inference, as implied by Bioportal, was ever
>     discussed during
>      > the RDFS standardisation process; and if so, why the more restrictive
>      > definition for multiple domains and ranges was chosen.
>      >
>      > I suspect this question has a simple explanation in model theory,
>     which is
>      > why I also copy Pat.
>      >
>
>     I recall this was discussed in the 2000-2004 RDF working group, or
>     at least
>     among some members of the working group at that time.
>
>     A concern here is for logical monotonicity - the introduction of new
>     knowledge
>     cannot invalidate existing knowledge, otherwise how can one know for
>     sure that
>     anything you think you know is actually true in a context that invokes
>     open-world semantics?
>
>     There are alternative models (e.g. default reasoning), but in order
>     to draw firm
>     confusions they require assuming that one has a complete set of
>     assertions (i.e.
>     no more can be added).
>
>     Also from the 2000-2004 RDF working group (which ran in parallel
>     with the first
>     OWL working group), the RDF list construct (aka
>     rdf:parseType="Collection") was
>     introduced so that (among other things) OWL could make closed
>     assertions, such
>     as owl:unionOf (see
>     https://www.w3.org/TR/2004/REC-owl-guide-20040210/#SetOperators).
>
>     #g
>     --
>
>
>

Received on Tuesday, 23 February 2016 14:33:55 UTC