Use of 'defined class' in OGMS

[GoogleCodeExporter]

Email submitted to IAO issue tracker by Chris Mungall:
-------------------------------------------------------
I think the defined-class vs universal dichotomy is confusing on
multiple different levels. It mixes separate orthogonal concerns -
information artefacts vs mind-independent things, primitive vs
complex, defined vs undefined, asserted vs inferred hierarchies.
Everyone I've met who has been exposed to the distinction seems
confused. A major part of the confusion is overriding the perfectly
simple term "defined class" to mean something other than "a class that
is defined". I would like the terminology to be changed.

The following OBI document is confused and will lead to further
confusion:

       http://obi-ontology.org/page/Defined_classes

It seems to suggest things like using protege to determine whether
something is a universal?

I think there is a potentially useful distinction that is being lost:
the distinction between classes representing patterns repeated in
nature (and thus of interest to science) and classes that are somewhat
arbitrary grouping classes, or grouping classes made for some specific
purpose. But this distinction is tricky, and should not be mixed with
other mundane yet important distinctions.

Let me try and clarify this with two diagrams. The first hierarchy is
of mind-dependent representational artifacts:

Representational Unit
       Class
               Defined Class
                       Defined Class, definition specified in formal language
                               Defined Class, definition specified in FOL
or subset of FOL
                                       Defined Class, definition specified
in OWL
                                       Defined Class, definition specified
in unrestricted FOL
                       Defined Class, definition specified in natural
language or semi-
controlled subset
               Undefined Class
                       Primitive Class
                       Potentially definable class, definition not
specified yet

This isn't necessarily pairwise disjoint (e.g. BFO classes may have
NL, FOL and OWL definitions), nor is it exhaustive (we're not
concerned with extensional vs intensional definitions, or with
partially specified definitions, this is just for illustration). These
are all within the scope of the IAO, and the distinctions made here
are all of some relevance to ontology construction, but of little
relevance outside this field. Hopefully these are reasonably boring
and uncontroversial, and don't involve taking a philosophical stance
about reality or science. There is one point of terminological
confusion, in that "defined class" means "class that has owl
equivalence axiom" to Protege users; however, I think it's important
to recognize other ways of specifying definitions, and to use the term
in the generic sense, and qualify it for added specificity if necessary.

The second hierarchy is of mind-independent entities. Please _don't_
think of this an an ontology or meta-ontology, it's just a diagram to
attempt to clarify some things:

Representation-independent entity
       Instance
       Instantiation pattern
               Universal
               Arbitrary pattern

I'm taking the realist position here whereby universals/types exist,
as this is consistent with BFO. Anti-realists may scoff at the notion
of a universal existing outside our minds, but hopefully they will at
least go along with the notion that there are some patterns in nature
that arise through scientific laws. Perhaps not - but the very fact
that this is contentious is one of the reasons we must separate these
two hierarchies.

The implicitly exclusive disjunction "defined class vs universal"
mixes these two hierarchies and conflates numerous unstated
assumptions about what kinds of things in hierarchy 1 can represent
what kinds of things in hierarchy 2.

By separating these hierarchies we can explore certain questions more
coherently:

* Is any undefined class necessarily intended to represent a universal?
* Should reference ontologies contain only classes that represent
universals, or can we have classes that are arbitrary groupings?
* Are universals definable? If so, are there constraints or guidelines
(e.g. genus-differentia style positive conjunction of a named class
and 1 or more differentia)?
* Does the single asserted inheritance hierarchy dictum apply only to
classes representing universals? what about inferred?

Different people will have different opinions on these - others may
think these are all irrelevant points of dogma. These discussions have
been happening, and the confusions arising from these discussions are
clouding the very simple, mundane yet important notion of  "defined
class". I suggest we reserve the term "defined class" to mean "a class
that has been defined", as this is how it is understood by the
majority of people outside this esoteric discussion -- and it's
completely obvious!! If we wish to draw separate orthogonal
distinctions then we should make these distinctions apparent in the
name and not pollute existing terminology. For example, something like
"universal class" vs "grouping class" or "arbitrary grouping class".

This all seems fairly obvious, sorry for belaboring the point.

One remaining difficultly is that "defined class" has been pressed
into action to cover cases where a BFO representation is non-obvious.
For example, OGMS places "phenotype" under defined class and there has
been discussion on using "defined class"es to represent "qualities of
processes" such as heartbeats. I wouldn't say these are arbitrary, and
I think any theory of universals that excludes phenotypes or
heartbeats would not be much a theory. I believe that ultimately we'll
have a way of doing justice to these in BFO that renders them as
universals, but this is a bit harder than the simple terminological
suggestion above.

Original issue reported on code.google.com by albertgo...@gmail.com on 10 Dec 2009 at 8:16

[GoogleCodeExporter]

I agree with Chris that using 'defined class' in the OGMS hierarchy is bound to
confuse.  I propose we use a better term: 

'nonuniversal class' to denote a class that is not a universal in the BFO
sense...i.e., not repeatable in nature

OR, perhaps

'named class' to denote a class that science gives a name to, but whose status 
as a
universal has not been determined

Original comment by albertgo...@gmail.com on 10 Dec 2009 at 8:20

[GoogleCodeExporter]

Original comment by albertgo...@gmail.com on 10 Dec 2009 at 8:20

• Added labels: Priority-High
• Removed labels: Priority-Medium

[GoogleCodeExporter]

We discussed this on the OGMS call, and although I dropped off before the 
discussion
ended, I just want to note that in the OWL world, the word 'class' refers often 
to
the representational unit, and not what is represented in reality.

I think any term we come up with that includes the word 'class' has great 
potential
to cause the same problems as 'defined class'.

Original comment by hoga...@gmail.com on 6 Jan 2010 at 7:10