On 3/10/2015 1:31 PM, Simon Spero wrote:
There are also named individuals that can appear in the T-Box - SH_O_IN(D).
There are some privileged individuals, such as the earth and the sun,
which are essential for defining geographical coordinates, times, days,
nights, etc. For any country X, it's impossible to specify the laws
of X without referring to X and some named entities in X. And for any
business Y, the business types and rules used for Y will normally make
many references to Y and some named entities in Y.
There can also be anonymous individuals as annotation values. (This
triggered a horrible bug in the OWLAPI when used with punning.)
Yes. And my major complaint about OWL is that it should be called
*An* Ontology Language, not *The* Ontology Language. Some things we
know about the currently popular languages: (a) they have changed
considerably over the past 10 years, (b) they will change even more
over the next 10 years, and (c) there are and will be many more
languages that will have to interoperate with them.
For some perspective on the history of interoperable systems and
proposed standards for them, see http://www.jfsowa.com/ikl
General principle: Flexible guidelines for ontology design are
useful. But rigid standards will be obsolete as soon as they're
written. I agree with the comments below by Pat Hayes.
John
-----Original Message-----
From: Pat Hayes [mailto:phayes@xxxxxxx]
Sent: Sunday, March 08, 2015 2:31 PM
To: Anthony Mallia
Cc: David Booth; public-semweb-lifesci@xxxxxx; HL7 ITS
Subject: Re: Proposed RDF FHIR syntax feedback
Comments in-line:
On Mar 8, 2015, at 9:00 AM, Anthony Mallia <amallia@xxxxxxxxxxxxx> wrote:
> David,
> I believe that this is an important aspect to distinguish between the
type or TBox and the instance or ABox. A simple justification is that
they come from different authorities (and end points) - HL7 or an EHR
system.
If there is any other reason to distinguish them, please list as many of
them as you can. If this is the only reason, I would strongly suggest
that it is not a sufficient reason for introducing this rigid
distinction into the foundation. It would be better to provide a
mechanism to allow the kind of originating authority to be specified
explicitly. The question to ask is, what utility in actual processing
will arise from having this distinction rigidly enforced? The problems
it (artificially) introduces is that it makes most OWL2 ontologies
unclassifiable, since many of them contain both class and instance data:
in fact, OWL2 punning makes this very distinction rather hard to detect,
since a class in OWL 2 may itself be an instance; and it forces users to
make a needless classification decision which may give rise to errors
and difficulties in processing.
> However I would strongly recommend that we DO NOT REDEFINE Ontology
from its definition in the W3C specs - this will cause major confusion.
> Here is the extract from OWL2:
> "OWL 2 ontologies provide classes, properties, individuals, and data
values and are stored as Semantic Web documents. OWL 2 ontologies can be
used along with information written in RDF, and OWL 2 ontologies
themselves are primarily exchanged as RDF documents."
That defines an OWL2 ontology. If you are planning to use other
representation languages, I would suggest adopting a wider definition of
the bare concept of 'ontology'. By the way, this topic - how to define
'ontology' - was discussed in depth for a year in the Ontolog forum. I
recommend reading
http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2007_Communique
and the surrounding discussions before coming to a decision.
> So I am recommending two subtypes of Ontology :
> INSTANCE ONTOLOGY (INSTANCE for short) contains Individuals, their
Property assertions and their data values but may refer to contents of
MODEL(s)
I think you mean it contains individual *names*, right?
When you say 'may refer to', what distinction are you making between
'refer to' and 'contain'? Do you mean it will not contain the
*definitions* of the classes, etc.? But there is no concept of
'definition' in the RDF/OWL world.
> MODEL ONTOLOGY (MODEL for short) contains Classes, ObjectProperties,
DataProperties and Datatypes
And what will you do with something which contains large amounts of
instance data, described using a mixture of vocabulary from a number of
other ontologies and a small number of class and property definitions
local to it? Because this is, if anything, the normal situation in
Web-based ontology work.
> INSTANCE and MODEL are disjoint
Which, if enforced, is going to create errors and blocks to processing
for no functional reason. Why do this? It is a bad design decision to
introduce distinctions that have no utility other than to be enforced
and generate error messages. If this is a genuine type distinction, then
you should be able to say what reasons there are for a processor to know
what type an ontology is. How will an INSTANCE be processed differently
from a MODEL?
> but there can be Ontologies (neither of these subtypes) which combine
them through merge or import and would be used for reasoning.
> It should not be necessary to separate these two by MIME type - they
will be handled quite differently e.g. import statements will know
exactly what they are trying to do.
importing is completely transparent to this distinction. Both of them
(and any hybrids) will be imported in the same way using the same
mechanisms. This is part of the RDF/OWL design.
Pat Hayes
_______________________________________________________