Re: SUO: Semantic interoperability
And of course I agree with you, John. As discussed in our FOIS-01 paper
"Ontological Engineering for B2B E-Commerce" (Obrst, Wray, Liu), we
found our greatest technical challenge last year at the B2B company
VerticalNet (I created a dept. of ontological engineering and we built
ontologies in the product and service space) to be the semantic mapping
issue (a portion of "semantic interoperability"). Unfortunately, mostly
what one must map to are very ill-defined taxonomies and classification
systems, and here I assume it is a mapping to a well-defined set of
"reference ontologies" as we did, just so you can try to make reasonable
inferences, have a baseline reasonable semantics. The semantics you
"preserve" is sometimes very local, i.e., portions of the
taxonomies/classification systems. Mostly everything is inconsistent and
unsound, but yet one must map to these (different owners, different
standards, etc.), preserving what one can.
Aside: the EC OntoWeb effort has a SIG on Content Standards. I urge many
of the folks here to look into affiliating:
http://www.ladseb.pd.cnr.it/infor/ontology/OntoWeb/SIGContentStandards.html
A point I make is that most of the declarative and procedural reasoning
"engines"/applications in the world are unsound and inconsistent and yet
are probably useful for their local targeted reasoning, and will
probably have to be used for some time. If you are only concerned with
mapping logically well-defined ontologies, it is much easier: do they
have equivalent or overlapping (by how much?) formal models. There are
other issues, but in any case, this latter situation is not what exists
in the world now, by far. If one can wipe the slate clean and design
logically well-defined ontologies (perhaps something we can do here at
SUO) from the ground up, how much easier things would be.
Leo
"John F. Sowa" wrote:
>
> On the NCITS L8 mailing list, there was some discussion of how
> the term "semantic interoperability" might be defined. Someone
> suggested a defintion taken from a standards document, which I
> criticized as a big fuzzy cloud of words. I'm sending a copy
> of that note to the SUO list because such a definition is also
> necessary for defining conformance for any system that uses
> an ontology.
>
> John Sowa
> _______________________________________________________________________
>
> The definition of "semantic interoperability" is precisely the kind
> of problem for which logic is required. To define such a term,
> you need a formal definition of "implication". Once that term
> has been defined, you can define other notions in terms of it.
>
> For example, if a proposition p implies a proposition q, and q also
> implies p, then p and q are said to be "logically equivalent".
>
> The suggested definitions of interoperability, including ones taken
> from standards documents, are essentially big fuzzy clouds of words.
> I have no doubt that such fuzzy clouds are better than nothing
> for helping people get some idea of the meaning, but they cannot
> support computer systems that are truly interoperable for any
> useful application.
>
> Joe Christensen wrote:
>
> > .... The words in 11179/3 about (paraphrasing)
> > "ensuring that the receiver understands the data as intended by the
> > sender" are pretty good, and I'll use these as the default if there's
> > nothing better.
>
> Using the notion of implication, I would formalize this definition
> in the following way (this is English commentary that could be
> translated directly into a version of logic, such as KIF or CGs):
>
> A sender's system S is _semantically operable_ with a receiver's
> system R if and only if the follow condition holds for any data p
> that is transmitted from S to R:
>
> For every statement q that is implied by p on the system S,
> there is a statement q' on the system R that
>
> (1) is implied by p on the system R, and
>
> (2) is logically equivalent to q.
>
> It is possible that the receiver R may have more implications than
> the sender intended, but the receiver must at least be able to derive
> a logically equivalent implication for every implication of the sender's
> system.
>
> Example: The sending system S might be a simple file server that has
> no capability for deriving any implication other than the trivial
> one that p implies p. Such a system would, by definition, be
> semantically interoperable with any receiver. However, it would
> not be able to guarantee that two different receivers would be
> interoperable with one another. More conditions on the definition
> are required to ensure that possibility.
>
> More conditions would also be needed to define semantic
> interoperability between systems S and R that use different
> languages for making statements. That can be done, but it would
> require another notion of "truth preserving translation" between
> different languages.
>
> You can't define semantics without having a formal language of some
> kind. In most practical cases, that formal language is some
> programming language, and the only effective definition is to
> declare some particular implementation as the reference standard.
>
> Bottom line: If you want to have a definition that is independent
> of a particular implementation, you need logic or some sugar-coated
> substitute for logic, such as SQL or a controlled natural language.
>
> John Sowa
--
_____________________________________________
Dr. Leo Obrst The MITRE Corporation
mailto:lobrst@mitre.org Intelligent Information Management/Exploitation
Voice: 703-883-6770 7515 Colshire Drive, M/S W640
Fax: 703-883-1379 McLean, VA 22102-7508, USA