Evelyn & Thomas, Thanks for all your work. I'm starting to go through
the deliverables, and immediately noticed something that I think needs
discussion.
My eye landed on two descriptions:
1. Some EDM classes imply the use of properties. If the class is used
these properties should be used.
2. EDM provides a list of mandatory properties per classes. However,
there are some conditions on certain classes requiring the use of at
least one of the mandatory properties.
I think the term "class" here is not being used with the same definition
as is used in RDF or OWL, and therefore we should clarify what is meant.
In these examples, class seems to be used for what I would define as a
dsp:descriptionSet. AFAIK, there is no way to say in RDF or OWL "for
this Class there must be a,b,c properties." That is because the
properties allow you to infer the class of the subject, but Classes
cannot define the properties. Classes allow you to infer semantics
related to the subject, but are not structures within the data (in the
way that a table is a container of data in a database, or an XML
property is a container of values).
This is hard to explain, so I take the liberty of quoting two paragraphs
from a paper that will soon be published, authored by Tom Baker, Sean
Petiya, and myself, from the section explaining how RDF differs from
what has gone before in data processing. We will reiterate this at the
pre-conference next week:
"How properties are associated with classes in RDF, with RDF domains and
ranges, is radically different from how properties are associated with
classes in systems based on the Closed-World Assumption, such as
relational databases, XML repositories, and object-oriented systems.
Closed-world systems prescribe templates to which data must conform,
with required structures and integrity constraints that can be used to
flag errors in the data. In such a prescriptive system, for example,
saying that “plays game” is a property of the class Athlete typically
implies that instances of the Athlete class can (or must) have a value
for “plays game.” Not declaring “plays game” as a property associated
with the class Athlete typically means that members of that class may
not be described with the property.
In RDF, domains and ranges are not used to bind a set of properties to a
class. Indeed, an OWL property cannot be limited for use only with
members of a specific class. Rather, RDF domains and ranges enable
inferences, such that if a “plays game” property is used to describe a
member of the class Person, a reasoner will simply state that the person
is also a member of the class Athlete. Where a closed-world system will
signal an error if a property is used in a context not intended by its
maintainers, an RDF reasoner will simply infer additional information.
An RDF reasoner may flag logical inconsistencies, for example in a
triple that says “X isDifferentFrom X,” however it will not validate
that data against a template of integrity constraints."
I do think that the DSP's description set provides something that, at
least conceptually, does not exist in RDF or OWL, and your use cases are
examples of that, but not of classes in the RDF/OWL sense. However, it
*may* be possible to get the same result with OWL or SPARQL. I believe
that the underlying meaning in OWL would be:
--for "thing" of type A, a propertyB predicate with "thing" as its
subject must be present in the set of triples
In fact, this would be true even if propertyB does not have a domain of
A. And that's where I see the difference between RDF and, say, OO or XML
-- class semantics are informative but not binding.
So the wording could be something like:
1. Certain EDM properties are required when there is a subject that is
an instance of a particular class. (?? other suggestions welcome)
But with the knowledge that including *any* property that is an instance
of the class results in an inference that the subject is a member of
that class. This is an area worth exploring with our test records.
kc
On 10/1/14, 7:15 AM, Evelyn Dröge wrote:
> Dear all,
>
> a draft of the first deliverable about case studies, use cases and
> requirements is now on our wiki:
>
> http://wiki.dublincore.org/index.php/RDF_Application_Profiles/UCR_Deliverable
>
>
> If we have reused your texts from the wiki or the database, you were
> added as creator. We have not added your email addresses but you can add
> them yourself if you are okay about having them on the wiki.
>
> We are looking forward to receive your feedback on the deliverable until
> November 3, 2014.
>
> Thank you very much in advance!
>
> Best,
>
> Evelyn and Thomas
>
--
Karen Coyle
[log in to unmask] http://kcoyle.net
m: 1-510-435-8234
skype: kcoylenet/+1-510-984-3600
|