[Alan Rector]

Gary, All

(I'm having trouble posting to the group from my usual email address since it changed.  I'm trying this while it is sorted out.)

Gary

Alan,

Thanks for kicking off some discussion of ontology "definition" with some apt issues.

One thing that I wonder about in your discussion is that you seem to talk about ontologies as a special kind of knowledge representation.  I tend to think of universals and generalizations, (empirically arrived at?) as you call them, as forms of knowledge (something understood epistemologically) which can be expressed as a knowledge representation (using a KR language).

Is that a useful distinction?

Gary Berg-Cross, Ph.D. 
gberg...@gmail.com    
​​​http://ontolog.cim3.net/cgi-bin/wiki.pl?GaryBergCross
Member, Ontolog Board of Trustees
Independent Consultant
Potomac, MD
240-426-0770

On Tue, Aug 2, 2016 at 9:43 AM, Alan Rector <Alan....@manchester.ac.uk> wrote:
All

I’ve been quiet and disengaged for some time for various reasons. 

I think this is an important effort, and important way to present ontology to the outside world and establish its appropriate role with respect to other formalisms and approaches in applications.  For that we need to be clear about what an “ontology” is and is not.

A few quick thoughts...

There is widespread confusion amongst the various sorts of artefacts that are sometimes called “ontologies” by various authors - which range from formal logical representations to data base models to frame systems to general logical theories.  I presume that the intention is that ontologies be confined to representations of universal entities and their necessary characteristics , what I sometimes term “ontology (narrow sense)” in the spirit of the definition given.   I am not convinced that the combination of the definitions for “ontology” and “universal” as given will be sufficient for those not already familiar with the terms.

More importantly and in a bit more detail,,,

It seems to me we need to clarify and distinguish:

1)    Ontologies and more general forms of knowledge representation.  All too frequently “ontology” is used almost as a synonym for knowledge representation, in which case the usefulness of the label for the part and/or kind of knowledge representation concerning universals is lost.  This is the heart of my concern about improving  I think the definition of "Ontology” needs clarifying.

2)     An agreed pair of terms for representations that are intended to be a) universal - either by definition or otherwise - and therefore indefeasible vs b) statements of what is generally true but defeasible - for which I would propose the label “Generalizations”.

3)      We need to be a able to say clearly what is, and is not, included in an “ontology” in our intended sense.  I would expect an “ontology”, in our sense, to include only universal statements and to exclude defeasible “Generalisations”. (Otherwise, how do we distinguish “ontology” from “knowledge representation” more generally?  For a given knowledge base, how do we say what is “ontology” and what is not?)

3)     Between the abstract “ontology” and its representation/implementation in any particular formalism.  Without that, how can we discuss questions such as the appropriate formalism for representing a particular ontology?  Or alternatively, a statement of what it means for two artefacts to be equivalent even though they may be represented in different formalisms.

4)     Between the different families of computational logic formalisms / inference mechanisms.  In particular we need to differentiate - and we need a better terminology for - a)  theorem proving axiom-based formalisms such as FoL, OWL-DL, DLs, CGs, etc. based on model theoretic semantics and the open-world assumption and b) database and logic programming formalisms using closed world reasoning mechanisms amongst a given set of symbols in a given model. 

Closely related,  would also be useful to have a better general term for the class of semantics of logic-programming and other closed world systems that provide proofs in a specific model/world to contrast them with model-theoretic semantics which provide proofs as to what is possible, or more precisely impossible, in any model/world.   (I know of names for specific methods - e.g. linear resolution - but can find no term for the general class of methods.  Perhaps it’s just my ignorance)

These may not be “core terms”, but they are central to many of the misunderstandings around “ontologies”.

5)  Query languages against a representation vs the representation itself - e.g. SPARQL queries vs OWL class definitions.  To be able to explain why “not classified under MyClass in this ontology” - something that can only be determined by a query against this representation - and the class of all things in the class "NOT MyClass".

Regards

Alan
------------------
Alan Rector
Professor of Medical Informatics (emeritus)
School of Computer Science
University of Manchester
Manchester M13 9PL, UK
www.cs.man.ac.uk/~rector

Please note new email address:
alan....@manchester.ac.uk