Re: Handling multiple rdfs:ranges

Reto,


On 23/02/2016 18:12, Reto Gmür wrote:
>
>
> On Tue, Feb 23, 2016, at 17:41, Peter F. Patel-Schneider wrote:
>> On 02/23/2016 08:24 AM, Reto Gmür wrote:
>>> On Tue, Feb 23, 2016, at 17:05, Peter F. Patel-Schneider wrote:
>>>> On 02/23/2016 07:31 AM, Reto Gmür wrote:
>>>>> [...]
>>>>>
>>>>> Granted, the semantics of :rangeIncludes are very weak (under OWA) but
>>>>> the fact that you can create contradictions with it shows that it's not
>>>>> completely meaningless.
>>>>>
>>>>> ex:prop1 s:rangeIncludes :Cat .
>>>>> :Cat owl:disjointWith :Dog .
>>>>> ex:prop1 owl:range :Dog .
>>>>>
>>>>> The above graph evaluates to false in every possible world, this is not
>>>>> the case if you omit any of the 3 triples, this shows that
>>>>> `s:rangeIncludes` is not a meaningless decoration.
>>>>>
>>>>> Reto
>>>>
>>>> I don't think that this follows from the semantics of :rangeIncludes,
>>>> even if
>>>> you augment schema.org semantics with disjointness.
>>>
>>> In the example I also used "owl:range" to create what I thought is a
>>> contradiction.
>>
>> OK, but I saw that.  (I actually missed that there are no values for
>> ex:prop1.
>>   Without any such values you don't get a contradiction even if you made
>>   both
>> of the ranges be OWL ranges, and used OWL semantics.)
> You're right, forgot [] ex:prop1 [].

Even with this, there is no inconsistency. The above statement plus the 
other statements you mentioned before say that there is something in 
relation with another thing of type Dog, and not of type Cat, via 
property ex:prop1. Where is the inconsistency?

An inconsistency may have arisen in two cases:

1) you explicitly say that ex:prop1 *only* rangeIncludes :Cat (which you 
did not). This can be done with some OWL constructs that go beyond OWL DL:

ex:prop1  a  [
    a  owl:Restriction;
    owl:onProperty  s:rangeIncludes;
    owl:allValuesFrom  [ owl:oneOf ( :Cat ) ]
] .

It is very unlikely that anyone will use schema.org in this way, and 
many OWL reasoners would have problems with this.

2) you interpret s:rangeIncludes in a different way by assuming that, if:

?p  s:rangeIncluldes  ?c .

holds, then there exist two resources ?r1, ?r2 such that:

?r1  a  ?c .
?r2  ?p  ?r2 .

Such interpretation is admissible but nothing in the spec indicates that 
it is correct.


However, if you use a heuristic saying that local data is complete, then 
I can assume that, in your example, :Cat is the only rangeIncluded class 
for ex:prop1, so I can detect an inconsistency as you said.


I would encourage people with a good KR / logic / model theory 
background to read Peter's paper on the subject of schema.org semantics:

Peter F. Patel-Schneider: Analyzing Schema.org. In Proc. of ISWC 2014: 
261-276. 
http://iswc2014.semanticweb.org/raw.githubusercontent.com/lidingpku/iswc2014/master/paper/87960257-analyzing-schemaorg.pdf?raw=true


--AZ


>
>>
>>>> Perhaps one could also count the documentation of
>>>> rangeIncludes as authoritative as well.  So from
>>>> https://schema.org/rangeIncludes, rangeIncludes "[r]elates a property to
>>>> a
>>>> class that constitutes (one of) the expected type(s) for values of the
>>>> property" would also be part of the semantics of schema.org ranges.
>>>
>>> I considered only this definition. And based on that I still think there
>>> is a contradiction, if the owl:range of a property excludes :Cat (which
>>> is expressed with the statements using owl-properties), :Cat cannot at
>>> the same time "be (one of) the expected type(s) for values of the
>>> property".
>>
>> Here is where we differ.  In my view, adding "expected" weakens the
>> statement considerably.
>>
> Right, "one of the possible types" was my interpretation, but something
> may be considered as "expected" even if there is only one irrational
> entity expecting it.
>
>> Without any official formal semantics for schema.org or other guidance
>> from
>> the schema.org people we are reduced to considering the meaning of
>> English
>> phrases on the schema.org website.
>
> Could it be triples all the way down? Doesn't the justification chain
> typically ends at some definitions in natural language?
>
>> Worse, the phrases used there aregenerally quite informal.
>
> This makes it difficult indeed.
>
> Reto
>
>
>
>

Received on Tuesday, 23 February 2016 19:06:13 UTC