Where to start (was: Ontology for Dummies)

[Sean Bechhofer]

On Wed, 2 Oct 2002, Tony Linde wrote:

> Welcome Steve.
>
> I don't want to spark an OntoWar, but could people (participants and
> lurkers alike) say if there is any significant difference between a
> topic map approach to ontologies and an OWL (or its ancestors') language
> approach.

Not sure if I qualify as lurker or participanet, but here goes! I'd
certainly like to avoid an OntoWar too -- so it's perhaps worth saying
that the following is purely *my* opinion and should not necessarily be
taken as representative of the entire DAML+OIL/OWL community. There, that
should do it :-).

Topic Maps and OWL (hereafter referred to as TM and OWL to save
typing) *are* different. This is not to say that one is "better" than
the other though -- they are *different*, and are suitable for
different purposes. TMs provide a kind of model of "back of the book
indexes", a way of indexing particular occurrences of topics or
subjects. In my opinion, they are not a framework for representing
ontologies (** see below). Again, in the interests of world peace and
harmony, this is not intended as a diss or put-down of TMs -- there
are other things that TMs will do better than an approach like OWL.

OWL is a language and framework for representing ontological knowledge
and information about the way that the world is structured and fits
together. As such it has a well-defined semantics (in terms of a model
theory) that allows us to know precisely what we mean when we use the
constructors in the language to define a particular class of
individuals.

Turning back to an old example from KR, what does the following mean?

   -------------             ---------
   | telephone |--- colour ---| black |
   -------------             ---------

o Telephones are black?
o All telephones are black?
o Telephones can be black?
o There is a telephone which is black?

The crucial point with OWL is that we have a well-defined notion of
what it means when we say:

telephone -> all colour black

i.e. all telephones can *only* have the colour black.

Drawing comparisons between TMs and RDF is not the same as drawing
comparisons between TMs and OWL. The language that OWL provides for
describing the domain is significantly richer than that provided by
RDF(S). RDF(S) effectively provides us with a taxonomic hierarchy and some
relationships between classes. We can't actually *define* things or talk
about, for example, the class of planets that have particular properties
(or at least not in a way that then allows us to draw any *inferences*
from that information, which is one of the key aspects of OWL). There's
very little inference that can be drawn, other than computing the
transitive closure of the hierarchies (i.e. running up and down the
subClassOf relationship).

As one of my colleagues has pointed out, it's relatively easy in all
these representations to put information *in*, but in order to get
anything other than non-trivial information *out*, you need the
semantics.

(**) Of course, we can always get into a discussion about what an
ontology is or isn't, which could go on for a long time :-). For the
purposes of this discussion though, I'm claiming that an ontology is
more than just a taxonomy or hierarchy, but has additional information
about the properties or definitions of the classes that you're talking
about. This is of course, not to dismiss taxonomies and simple
controlled vocabularies -- one can get a long way with such things. If
you look at the experiences of the medical and biology communities,
however, maintaining and constructing such vocabularies becomes
problematic when the models get large. This is where an approach using
a richer language and computational support can benefit (e.g. see the
experiences of GALEN (http://www.opengalen.org), and more recently
using DAML+OIL, GONG (http://gong.man.ac.uk)).

Cheers,

	Sean

-- 
Sean Bechhofer
seanb at cs.man.ac.uk
http://www.cs.man.ac.uk/~seanb