Re: REQUEST: survey of available ontologies...

Folks,

In the interests of reducing the volume, I'll combine my comments to
several of the postings into one note.  Unfortunately, many of those
postings came from different lists, so I'm doing some cross-posting.
I promise that this is my last reply to this thread.

Frank van Harmelen wrote:

 > I guess you refer to diagrams such as [1].

 > This is only meant to convey that RDF is the syntactic carrier for
 > ontology languages and other logical formalisms (and XML is in turn
 > the syntactic carrier for RDF).

 > I don't think it was ever meant to imply that logic-based analysis
 > of a domain would not be needed before using RDF/RDF Schema as a
 > notation.

 > The picture is only a rather matter-of-fact statement about the
 > technical infrastructure for the Semantic Web, not a deep
 > methodological issue.

I agree that is one interpretation.  But I also recall a talk by
Eric Miller, in which he commented that the layers get less well
defined as you go up.  That annoys people like me who believe
that classical FOL is the best defined of all notations on earth.

I think that any layer-cake diagram would be misleading in some way,
since logic is needed at the lowest level to define all the other
notations, and it is also needed at the level of tools that do
reasoning -- deduction, induction, and abduction.

David Sallach wrote:

 > Are 1) logic and 2) ontology as separable as the above subdivision
 > suggests?  That is, while I agree that formal structure and rules of
 > inference are highly desirable, it is less clear that logic itself
 > has reached a final, stable form.  Rather, while better defined than
 > ontology, logic may still be evolving as well and, in fact, may
 > coevolve with putative ontologies, in which case, the initial
 > subdivision may be misleading.

I admit that any definition that is simple enough to be understandable
must leave out many qualifications that could make it misleading.
Later in the book, I qualify the distinctions by listing the
characteristics of various notations that are called logics, and
note that many of them, especially variants of temporal logics,
really incorporate major ontological assumptions.

 > The particular initiative I have in mind is situation theory (and
 > related frameworks such as relevant logic and dynamic logic).  Because
 > natural language is one of the primary domains of situation theory,
 > it addresses many of the issues discusssed in the Program Semantics
 > thread.  However, it also attempts to extend predicate calculus by
 > adding existential as well as syntactic assertions and constraints.

I agree that those formalisms mix ontological assumptions with the
logical formalism.  I also mention music notation in Ch. 1 of the book
as a classic example of a notation that mixes a simple version of logic
(with only conjunction and an implicit existential quantifier) with
an ontology for sequence, duration, pitch, loudness, and concurrency.
Such mixtures are very useful for many purposes, but I believe we
should determine exactly where the logic stops and the ontology begins.

 > We cannot yet know the ultimate contributions of these movements,
 > but is it not possible that some vexing problems may be addressed
 > most effectively by the emergence of novel ontologies combined with
 > innovative formalisms?

I am happy to endorse work on the development of novel notations that
mix logic and ontology in any way that the designers believe is useful.
And again I would cite music notation as an excellent example of the
genre.  But I believe that the resulting mixture should not be called
a "logic", but rather a language or notation that explicitly mixes
logic and ontology.

Bill Andersen wrote:

 > There are several relatively weak representation formalisms out there
 > that are being touted as suitable for building "ontologies", most
 > notably in the WWW/W3C communities.  XML/S is only one of those often
 > heard of in this connection.  The claim that they are sufficient for
 > ontology is implicit in the advertisement of this or that project,
 > written in these formalisms, as "ontology".

As I said before, I have no quarrel with people who design a notation
that may combine a limited logic with a special purpose ontology.
Music notation is an example, but no one claims that music notation
can solve more general problems of ontology.

 > These languages *are* logics of a sort.  Worse yet is that they pack
 > in all kinds of implicit philosophical assumptions that, in the
 > hundreds of pages of detailed specifications that come with these
 > beasts, get lost in the noise.  Nobody (probably even the language
 > designers) know they are there.

I agree.

 > But they (the philosophical assumptions) have real impact.
 > Necessarily existing properties?  No problem: OIL (or any DL) has 'em!
 > No theory of identity?  No problem: different name -> different
 > object!   Relational properties (e.g. "Italian")?  No problem: they
 > don't exist because they cause computational problems.
 >
 > As a friend of mine always says, let's not confuse activity with
 > progress.

To paraphrase Socrates, "An unexamined notation is not worth writing."

Jim Hendler wrote:

 > ...while this is true, it doesn't mean the logic must be explicit
 > in the language, just that their must be a formal model underlying
 > the language...

I agree.  But I believe the designers of the language should pay more
attention to these issues, instead of throwing the language over the
wall and hoping that somebody like Pat Hayes will catch them, clean
them up, and give them the good logic stamp of approval.

 > ... But John, you yourself know that there is a clear difference
 > between the encoding and the logic - so how can you make such a
 > ridiculous claim.  Of course there must be an underlying logic -- but
 > that logic doesn't need to be expressible on the web to be useful --
 > the underlying logic below DAML+OIL, for example, can be expressed in
 > KIF (see the axiomatic semantics [1] ) or Model Theory (see the
 > semantic model [2])

I know that and you know that.  But as Bill pointed out, there are
so many different notations with different underlying logical and
ontological assumptions that no one -- not their designers, their
users, or their reviewers -- knows how each of them is related to
each of the others.  Giving a model theoretic foundation for each
notation can prove that each one is consistent by itself.  But it
cannot show that using several of them together will be consistent,
meaningful, or usable for any practical application.

And your own reply to Bill illustrates the problem:

 > ...  When you write classical logic with all the usual symbols it is
 > meaningless scrawling on a piece of paper until we have the social
 > agreement about how the symbols map to mathematical concepts.

I agree.  Now imagine a few billion web pages with hundreds of XML
based languages scrawled across them.  Each one might have been well
designed by itself, and all of them may be processed by a common set
of XML tools.  But where is the social agreement that relates the XML
tags of one notation with its implicit logic and ontology to the XML
tags of another notation with a different set of implicit assumptions
about logic and ontology?

John Sowa

Received on Monday, 1 April 2002 22:31:39 UTC