Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: SUO: The Story So Far - Request for vote




At 4:13 PM -0800 10/3/01, Adam Pease wrote:
>Nicola,
>  Neither the approach of a single common ontology, or a catalog of 
>related ontologies has been proven to be the right approach, so 
>either or both may turn out to be valid.  However, I would say that 
>a single common ontology, to the degree practical, is the goal of 
>this group, and has been voted on when the PAR was approved.  I have 
>found that there can be disagreement on wording that I thought was 
>very clear, but the sections that seem to indicate this are (with my 
>emphasis by bracketing with '*'s):
>
>"This standard will specify *an* upper ontology that will enable computers..."
>
>"The SUO will play the role of a neutral interchange format whereby 
>owners of existing applications will be able to map existing data 
>elements just *once to a common ontology*."

Dear Adam,

	I agree that - strictly speaking - the former statement above 
(but not the latter!) is incompatible with a non-monolythic approach 
(just because I assume that *an* ontology should be described by a 
consistent logical theory). I think however that the practical goal 
of our enterprise would remain mainly the same if we adopt a broader 
perspective, replacing "an upper ontology" with something like "A 
framework of carefully crafted foundational ontologies". In practice, 
such a framework would be a set of logical theories ordered according 
to a theory-inclusion relationship, as proposed (I believe) by Chris 
Menzel. The formal structure would be similar to that of the Stanford 
Ontology Library, or to the ONIONS ontology. Differently from these 
approaches, the SUO would explicitly account for a (hopefully 
limited) set of well-documented ontological alternatives, to be 
selected by the various applications that want to "plug in" the SUO.

	Ideally, the SUO interface should offer the new user a "menu" 
of i) well-documented, ii) reasonably formalized, and iii) 
sufficiently agreed on ontological choices to choose from. On the 
basis of this choice, the interface would select the right set of SUO 
modules.

	In my opinion, this approach would be perfectly consistent 
with the latter statement above: the goal is to have a standardized 
semantic framework to be used to understand and relate each other the 
intended meaning of data elements, not so much to force data elements 
to commit to the same semantics!

	Moreover, the non-monolithic architecture is the only way to 
guarantee the high coverage of our initiative, which otherwise would 
necessarily exclude important players. For example, consider the 
community of so-called "linguistic" ontologies, explicitly 
"dismissed" by the recent Pat's messages... I believe it would be 
*essential*, for the purpose of the SUO, to be able to include 
linguistically-motivated domain ontologies in the SUO framework. In 
some cases, this may imply to adopt ontological commitments that are 
just incompatible with those useful for, say, nuclear physics 
ontology...

>I would say that Cyc falls somewhere in the middle on this issue and 
>I suspect that will be the result for the SUO as well - a largely 
>common ontology with a context mechanism that allows for a few 
>different, although compatible, approaches in limited areas.

I believe that an alternative could be a "minimal" common ontology 
with a few main branches, which in turn may have more specialized 
branches in limited areas. Of course, the less branches the better.

In practice, my proposal is as follows:

1. Let's carefully (and painfully) distinguish between axioms and 
primitives that achieve (reasonably large) consensus and 
axioms/primitives that do not;

2. When there is no consensus, let's isolate the alternatives and ask 
the supporters to i) explain, ii) motivate and iii) formalize both of 
them;

3. If enough explanation/motivation/formalization is achieved (up to 
a certain minimal degree) for both alternatives, then:
    3.1 create a new branch in the ontology library
    3.2 go on specializing the two branches in parallel, depending on 
the interest they
        receive.

4. Else proceed by considering only the alternative that was best 
explained, motivated, and formalized.

Note that this proposal obviously reflects a top-down approach, but I 
believe it can be adapted also to the middle-out approach advocated 
by Mike Uschold, as long as we are ready to create a new upper-level 
branch while generalizing from a certain middle-level view that turns 
out to be incompatible with the previously developed ontology.

-- Nicola



  ---------------------------------

Nicola Guarino
National Research Council	phone: +39 O49 8295751
LADSEB-CNR		fax:   +39 O49 8295763
Corso Stati Uniti, 4		email: Nicola.Guarino@ladseb.pd.cnr.it
I-35127 Padova
Italy

http://www.ladseb.pd.cnr.it/infor/ontology/ontology.html
(***updated 22/2/2001 ***)