Re: related measurements (MATCHING) for property names

Thank you Kalp to share your interesting question!

Dear Simon and Mike,

Your answers show that there is a need for a document clarifying the 
possible equivalence relations between "subjects".
With a precise "semantic" for each (semantic meaning for me "what 
automated processing a machine can make based on that relation?").

* For instance, "skos:narrower" is meaning "Concepts linked can be used 
to automatically extend a search equation"
* I am often wondering about "skos:related": does it means "Concepts 
linked should be presented to the user as alternate search" (i.e. what 
is the desired automated behavior?)
* And "skos:narrowMatch": "Concepts linked can be used to automatically 
extend a search equation if they are reachable within a reasonable 
computing effort"
* owl:sameAs has a very powerful semantic (consider the other subject as 
a perfect alias for every computing purposes)...

Do you have a small set of references for such a recapitulation?

Have a nice day!

Christophe
@ChristopheDupri

Le 5/07/2012 23:33, Simon Spero a écrit :
> On Thu, Jul 5, 2012 at 2:43 PM, Kalpa Gunaratna <kalpa@knoesis.org 
> <mailto:kalpa@knoesis.org>> wrote:
>
>     Hi,
>        I know that SKOS has several matching properties for concepts
>     in ontologies like |skos:closeMatch|, |skos:broadMatch|,
>     |skos:narrowMatch|and |skos:relatedMatch||. I want to get some
>     notion to say that ||two property names in two ontologies relate
>     or closely match each other. Since SKOS is defined for concepts, I
>     think I can't use it for my purpose here. For example, I want to
>     use something like 'SKOS:closeMatch" for two property names if
>     they mean the same in two datasets. I want to know whether there
>     is work going on regarding this issue and if so point me to a
>     resource so that I can use it for my purpose here. Thank you.|
>
>
> There are several possible answers depending on what is intended by 
> (a)"two property names"/ (b)"mean[ing] the same"/ (c) "in two datasets".
>
> The reading that seems most plausible to me is that:
>
> 1: There are two sets of individuals DS1 and DS2.   To simplify 
> things, we can let D1 and D2 be classes whose extensions are the 
> subjects of assertions in  DS1 and DS2   (c)
> 2: There are two predicates, P1 and P2       (a)
>
> 3: If two properties named by A and B,  mean the same thing in any and 
> all contexts, then the properties have  identical extensions  (forall 
> (?x ?y ) (iff (A ?x ?y) (B ?x ?y)).
>     If this is the case, then they are *owl:equivalentProperty*  . 
>  This does not mean that they are the /same/ property; it just means 
> that they have the same extension.  If the two properties /are/ really 
> the same property, then the properties are also *owl:sameAs*  .
>
> 4: If two properties named A and B  are not *owl:equivalentProperty *, 
> there must be some combination of subject and object where  A holds, 
> but B does not.
> If the properties do have the same extensions for classes D1 and D2, 
>  we can express this in CLIF as:
>        (iff  (D1or2 ?x) (or (D1 ?x) (D2 ?x))
>        (forall ((?x D1or2)  ?y) (iff  (A ?x ?y) (B ?x ?y)))
>
> In OWL we are a bit more restricted: however,  we can get the same net 
> result we create a new property C which is an rdfs:subProperty of 
>  both A and of  B, and whose rdfs:domain is D1 or  D2, and replace all 
> assertions of A and B in DS1 and DS2 with assertions of C.   From any 
> assertion ?x C ?y, we can infer ?x A ?y and ?x B ?y.  We can also 
> infer that ?x type D1 and ?x type D2 cannot both be false.
>
> Simon
>