Re: SUO: General Design
Kenneth,
Your experiences with existing tools and approaches gets
into some fundamental issues about how current ontology
tools -- and ontologies themselves -- are being designed.
And unfortunately, as I have been saying for over a
dozen years, the current designs are not only wrong,
they are totally, fundamentally, and disastrously wrong.
The single biggest mistake is the idea that an ontology
should be designed from the top down. I have no quarrel
with the idea of starting at a single node, which is
variously called T, Entity, Universal, Thing, or some
other innocuous name. My objections, however, start
at the level immediately below T and continue down
to the levels that people really want to use.
Whenever I make comments like that to certain people,
such as Doug Lenat, he is very quick to add that the
top levels of Cyc are not important. The really
interesting work, he says, is at the intermediate
and lower levels. However, I maintain that the same
reasons why most top-level ontologies are bad are
also causing trouble at every level beneath the top.
At the opposite extreme from the so-called theoretically
designed ontologies based on principles, ground rules,
and philosophically approved assumptions is the world's
most widely-used, but not theoretically certified
ontology -- WordNet.
I will not defend the nondesign approach to building
WordNet other than to say that its lack of design saves it
from the single most disastrous mistake of the theoretical
ones: rigidity and inability to accommodate change.
Every theoretician who looks at WordNet says that its
top levels are bad, but that its lower levels are OK
because they can be aligned or mapped to his or her
pet ontology. But what they fail to observe is those
mappings are not exact equivalences.
WordNet's strength is its loose design with synsets (the
so-called synonym sets of terms) which are loose enough
that they can bend and sway to accommodate different
applications. When you map them to a rigidly defined
structure, you cannot get an exact equivalence. All you
have is a bending, swaying structure decorated with ropes
that tie the synsets to the rigid nodes of a fixed,
unbending structure.
On a calm day, the two structures might look as if
they were bound together. But when the first storm
or earthquake hits, the rigid structure collapses,
the ropes break, and the bending, swaying structure
creaks and groans, but stays upright -- at least
as upright as it can ever be.
Since 1987, I have been arguing that no ontology or
knowledge base can ever be adequate unless it comes
to grips with what I have called the "knowledge soup"
-- the loosely organized, semi-structured mix of
whatever people have in their heads. It certainly
does not look anything like a rigid ontology.
WordNet is probably a lot closer -- but still not
close enough. There is a lot more to say on this
topic, but I'll leave it for later.
John Sowa
______________________________________________________
Kenneth Fields wrote:
Greetings,
> I've been working with ontology building for a few weeks now,
> and think I have a basic foothold - just kidding - long way to go!
> I know some of you are 'Classes' yourself.
>
> I've compared the SUMO and Cyc ontologies - though I haven't
> been able to load the Cyc upper ontology (daml file) into Protege
> because
> of problems with the file itself - according to a daml verification.
> Thanks
> for providing a Protege project file for the SUO on the suo ieee web
> page.
>
> My problem is now of course building my micro and meso
> ontologies and suturing the ganglia carefully to the upper ontology.
> This is a daunting task at first glance. What were apparently neat
> discipline or topic areas have now become (happily) a major
> deconstruction effort.
>
> Take 'Sound' for example, my present task. I am trying to reorganize
> my computer music class to be OC (ontological correct) (just
> for fun). In other words - we begin at RadiatingSound and then
> might jump over to InformationBearingContent to look at techniques
> to visualize a sound wave. Then I need to have quick access to such
> classes as Hertz, Signal, Theory/Cybernetics, DSP, Russolo - Art of
> Noise,
> Hearing, Pattern, Phenomenology, Music, etc. Linking to resources,
> examples, images, sounds, or movies is a problem. What I need
> is a cross between Protege and Power Point :)
>
> My question is:
> 1. General - any advice from experience - how to construct from the
> bottom up. For example, in Protege, it isn't easy to reorganize the
> class structure by adding a super class or inserting a mid-level class.
> You can do this, but it doesn't change the class hierarchy, it will
> just insert
> the symbol for multiple superclasses. Optimally, I would like to make
> small projects and merge them into meso structures later. But then
> I might not have access to inherited class properties.
>
> 2. Here's my first candidate for a micro-ontology.
> 'Sample' -> aSample (instance)
> A Sample is the elemental atom of DigitalSound. It is a number in
> ComputerMemory. It is assembled into an Array in a ComputerBuffer
> from a DataFile (many different instances or Formats of DataFile).
> A Sample has one dimension - magnitude = 0...1. The magnitude
> is also limited in resolution depending on 16 or 24 Bit systems.
> The width (timeDuration) of the Sample is later dependent upon
> parameters of SamplingFrequency.
>
> So every noun and verb is important and requires a separate field
> trip to some branch of the ontology tree - which I am hoping
> will be the main 'feature' of this teaching method.
>
> Thanks for your comments and let me know if I am barking
> up the wrong ontology tree - listServes. You might refer me to a
> more appropriate discussion list for lower ontology building -
> downtown as it were.
>
> Dr. Kenneth Fields
> Professor Media Arts/Computer Music
> CEMC - China Center for Electronic Music
> Central Conservatory of Music
> 43 BaoJia Street
> XiCheng District
> Beijing, China, 100031