Re: Handling multiple rdfs:ranges

My paper has a lot of discussion of schema.org domains and ranges.  The paper
does not refer muuch to s:rangeIncludes (or s:rangeIncludes) as these IRIs are
not part of schema.org per se.  I view the discussion of domains and ranges in
the paper as relatively complete.

One problem with domains and ranges in 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.

My semantic treatment of schema.org is a mathematical construct.  I suppose
that my employer has some rights to it, but it has been described in the
scientific literature and is thus available for scientific inquiry.  As far as
I know, my treatment has had no effect within the schema.org group.  As far as
I know, reliably ingesting schema.org data requires a large amount of
background general purpose information (to handle strings as things) and a
complex cleaning pipeline (to handle systematic and point errors in the data).
 This makes using schema.org data much less desirable for me.  One goal of the
project to analyze schema.org data was to see how much schema.org data
required complex methods to ingest.

peter



On 02/23/2016 09:25 PM, Ross Horne wrote:
> Hi Peter and Simon,
> 
> Your ISWC'14 paper is the kind of systematic approach I'm looking for. I
> notice that the paper focusses on properties other than schema:rangeIncludes,
> but you mentioned in this thready that you had considered this property. Are
> your proposed semantics in the public domain, or should I just be patient and
> hold on for an extended version of the paper?
> 
> Simon's view is close to my intuition, if I understood correctly. There is no
> closed world; instead, every consumer on the Web of Data has their own current
> view of schema information they have explored so far (as represented by
> Simon's anonymous owl:unionOf for example). When new schema information is
> discovered the current view is refined (in a monotonic way). This is a quite
> type theoretic perspective on ontologies, where the "current view" is
> basically a "type environment".
> 
> Excuse me for not being up-to-date with the lingo: is "SDO sponsors
> validators" a technical term?
> 
> Kind regards,
> 
> Ross
> 
> 
> 
> On 24 February 2016 at 06:00, Peter F. Patel-Schneider <pfpschneider@gmail.com
> <mailto:pfpschneider@gmail.com>> wrote:
> 
>     Yes, this is one way to think of schema.org <http://schema.org> ranges.  I
>     have asked whether
>     something like this was this case, and didn't get any response.
> 
>     Other meanings for schema.org <http://schema.org> ranges might be more
>     likely.  For example, one
>     might consider schema.org <http://schema.org> ranges to be constraining,
>     i.e., that a property
>     s:p1 with definition
>       s:p1 s:rangeIncludes s:cr .
>     (and no other s:rangeIncludes triples) constrains valid inputs to have type
>     triples to s:cr or one of its subclasses for every node that is the value
>     of s:cr.
> 
>     Which is right?  At one time I was hoping that I could get a public answer,
>     but it was not to be.
> 
> 
>     It is definitely not the case that one can just think of schema.org
>     <http://schema.org> "triples"
>     as OWL axioms and facts.   The treatment of domains and ranges in
>     schema.org <http://schema.org>
>     mean that this simple transformation doesn't work.  Even doing some sort of
>     closure also doesn't work, because of strings as things.
> 
>     peter
> 
> 
>     On 02/23/2016 01:22 PM, Simon Spero wrote:
>     > The interpretation of rangeIncludes etc., becomes easier if one hand-waves in
>     > a simple temporal context .
>     >
>     > A canonical reference oracle (iming danbri) accepted the assertion that the
>     > set of rangeIncludes axioms could be considered closed for a given version of
>     > schema.org <http://schema.org> <http://schema.org>.
>     >
>     > The included ranges form an anonymous unionOf; the effective range is the
>     > conjunction of this anonymous range with all other range assertions applicable
>     > to property, whether through assertions or by inheritance.
>     >
>     > Inferences from assertions in a document using schema.org
>     <http://schema.org> <http://schema.org>
>     > semantics should be made with respect to the version of the schema that
>     > existed at the time the assertions were made.
>     > This behavior roughly corresponds to the behavior of the various sdo sponsors
>     > validators.
>     >
>     > This assumption allowed for relatively simple mapping to OWL (literal types
>     > were just converted to classes, with magic boxing/unboxing).
>     >
>     > Generating named classes for the anonymous unions and computing the class
>     > hierarchy revealed a good bit of hidden structure, and also uncovered
>     > anomalies caused by errors.
>     >
>     > One interesting idiom that initially made no sense until it is explained is
>     > the use of ranges that are (Text or URL), where URL is a subclass of Text.
>     > This generally indicates an identifier of some kind, where the URL is in
>     > principle pointing to a named individual.
>     >
>     > What makes this interesting is that there is no ready way in OWL to restrict
>     > the range of an object property to be a named individual, since that
>     > distinction is purely syntactic. It's easy enough to sort of handle this
>     > poorly (checking for unacceptable anonymous individuals in input, generating
>     > different individual assertions, and discarding inferred anons in post).  It's
>     > difficult to handle this cleanly without bringing up a whole raft of UNA
>     > issues (and CWA issues if cardinality constraints are around).
>     >
>     > Simon
>     >
>     > On Feb 23, 2016 2:53 PM, "Peter F. Patel-Schneider" <pfpschneider@gmail.com <mailto:pfpschneider@gmail.com>
>     > <mailto:pfpschneider@gmail.com <mailto:pfpschneider@gmail.com>>> wrote:
>     >
>     >     On 02/23/2016 09:12 AM, Reto Gmür wrote:
>     >     >
>     >     [...]
>     >     >> Without any official formal semantics for schema.org <http://schema.org>
>     >     <http://schema.org> or other guidance
>     >     >> from
>     >     >> the schema.org <http://schema.org> <http://schema.org> people we
>     are reduced to considering
>     >     the meaning of
>     >     >> English
>     >     >> phrases on the schema.org <http://schema.org> <http://schema.org>
>     website.
>     >     >
>     >     > Could it be triples all the way down? Doesn't the justification chain
>     >     > typically ends at some definitions in natural language?
>     >
>     >     Well, maybe.  There is some stuff that has been machine-validated. 
>     (Which
>     >     then makes the basis some computer code, I guess.)
>     >
>     >     One big reason for formal semantics is to ground on something that
>     is quite
>     >     precise.  Grounding on simple model theories is useful, I think,
>     because there
>     >     is very little wiggle room left in the definitions and
>     constructions, even
>     >     though there is, as you say, still a natural language component that
>     has to be
>     >     considered even if the natural language is some language that
>     mathematicians
>     >     use to communicate with each other.
>     >
>     >     >> Worse, the phrases used there are generally quite informal.
>     >     >
>     >     > This makes it difficult indeed.
>     >     >
>     >     > Reto
>     >
>     >     peter
>     >
>     >
> 
>

Received on Wednesday, 24 February 2016 14:39:48 UTC