W3C home > Mailing lists > Public > semantic-web@w3.org > February 2016

Re: Handling multiple rdfs:ranges

From: Peter F. Patel-Schneider <pfpschneider@gmail.com>
Date: Thu, 25 Feb 2016 17:06:38 -0800
To: ross.horne@gmail.com
Cc: Simon Spero <sesuncedu@gmail.com>, Semantic Web <semantic-web@w3.org>, Reto Gmür <reto@wymiwyg.com>, Alan Jeffrey <ajeffrey@bell-labs.com>
Message-ID: <56CFA51E.8070304@gmail.com>
On 02/24/2016 08:42 PM, Ross Horne wrote:
> Hi Peter,
> 
> On 24 February 2016 at 22:39, Peter F. Patel-Schneider <pfpschneider@gmail.com
> <mailto:pfpschneider@gmail.com>> wrote:
> 
>     My paper has a lot of discussion of schema.org <http://schema.org> domains
>     and ranges.  The paper
>     does not refer much to s:rangeIncludes (or s:rangeIncludes) as these IRIs are
>     not part of schema.org <http://schema.org> per se.  I view the discussion
>     of domains and ranges in
>     the paper as relatively complete.
> 
> 
> I see. Your paper makes things much clearer, even if not official from
> schema.org <http://schema.org>. In schema.org <http://schema.org> ranges are
> already disjunctive.
>  
> 
>     One problem with domains and ranges in schema.org <http://schema.org> is
>     their behaviour when
>     extra domains and ranges are available. Constructs that are invalid in some
>     way without the extra domains and ranges are valid with them. This is, in some
>     sense, non-monotonic behaviour.
> 
> 
> I like the way you carefully say "in some sense, non-monotonic behaviour". In
> another sense, this is completely monotonic behaviour. In this other sense,
> properties are contravariant (just as inputs to functions or data sent over a
> channel). Thus to refine the type of a resource you strengthen the type, using
> intersection for example; but for properties you do the opposite, you weaken
> the type, using union for example. In this way, we can monotonically refine
> the schema information to accommodate both new explicit type information and
> new range triples.

I used the qualifier to indicate that I meant in a logical sense.  Perhaps I
should have been more explicit.

In a logical sense, and that's the one that should be used when looking at
schema.org as logical information, schema.org domains and ranges are
non-monotonic. Adding more input information results in fewer consequences.

You can consider schema.org markup to be something other than logical
information.   This opens up a lot of possibilities in how you can treat
schema.org markup.  However, it opens up a lot possibilities in how you have
to treat schema.org markup.

> Despite many helpful pointers, I am still unconvinced from this thread that
> there is practical rational behind the decision by the RDF Schema working
> group to make property domains and ranges conjunctive rather than disjunctive.
> The example given by AZ was unclear until he highlighted he was also
> considering female lizards;) However, on the other hand, there are scenarios
> where two people state that the range of a property is different type. E.g.,
> in DBpedia, dbp:subdivisionName is sometimes used with range int and other
> times used with URI dbp:Place:
> 
>   dbpedia:Rockmart,_Georgia dbp:subdivisionName 1872 .
>   dbpedia:Rockmart,_Georgia dbp:subdivisionName dbpedia:Polk_County,_Georgia .
> 
> This can easily happen with more than one person contributing data, as
> intended on the Web of Data. However, if both authors separately contributed
> respective schema information for the range, then under the conjunctive
> semantics any triple in which dbp:subdivisionName appears as the property is
> inconsistent. In contrast, under a disjunctive semantics both contributors
> datasets can be used together -- even without prior coordination -- as
> expected on the Web of Data. I am possibly opening a Pandora box here, by
> using an int along with a URI in this example; but similar weaker examples
> using URIs only and resulting in nonsensical conjunctions instead of full
> blown inconsistencies can be constructed.

One problem with this approach is when you want to use the information.  You
can't count on the fillers of a property to belong to the class (or datatype)
that you expected them to.  How, for example is an application supposed to
process subdivision information when it could be a place, or an int, or a
string, or a person, or ... anything?

> 
> Using owl:unionOf, the above problem can be resolved, but only with concious
> coordination between the two contributors and also by stepping slightly out of
> RDF Schema.
> 
> Of course, a standard is a standard, so I should say no more. However, if
> anyone does have a strong either practical or model theoretic case for the
> conjunctive handling of domains and ranges in the RDF Schema spec then I would
> be most interested to hear at any point.

The way to treat RDF information has been a subject of debate from the
beginning.  The current W3C standards generally treat RDF triples as
independent logical axioms.  Using this treatment biases how the rest of RDF
is set up.


> 
> Kind regards,
> 
> Ross
> 

peter
Received on Friday, 26 February 2016 01:07:10 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 19:49:52 UTC