ONT RE: Ontology case study
Dear Jim,
See some responses below.
Matthew West
Principal Consultant
Shell Information Technology International Limited
Shell Centre, London SE1 7NA, United Kingdom
Tel: +44 20 7934 4490 Other Tel: +44 7796 336538
Email: matthew.r.west@is.shell.com
Internet: http://www.shell.com
> -----Original Message-----
> From: Jim Farrugia [mailto:jim@spatial.maine.edu]
> Sent: 30 May 2002 14:37
> To: John F. Sowa
> Cc: West, Matthew R SITI-ITPSIE; Adam Pease; ontology@ieee.org
> Subject: RE: Ontology case study
>
>
>
> John, Adam, et al.,
>
> I am (still) trying to understand when and why one would use
> an ontology
> versus a data model for a given goal, such as answering
> queries. There are
> probably many answers; I'm just looking for some that I can
> understand.
>
>
> John said, in part, ...
>
> <snip>
>
> JS: > Although I agree with Matthew's claims for data models,
> I would like
> > to point out that there is no difference in principle between data
> > models and ontologies. Following are the correspondences:
> >
> > 1. Both have types and relations (AKA predicates), which are or
> > can be organized in a hierarchy of types and subtypes.
> >
> > 2. Both require definitions, although the DBMS models usually use
> > some less rigorous language, such as English. But as I have
> > insisted for a very long time, the definitions should be in a
> > structured, stylized, or controlled English (or other natural
> > language) that has a direct mapping to logic. If we used such
> > languages, the difference in appearance between the DB and AI
> > definitions would disappear.
> >
> > 3. There is no difference in principle between the DB constraints
> > and the AI axioms. In fact, the DB constraints,
> which are usually
> > stated in SQL can be translated directly (even automatically)
> > to any common notation for FOL, such as KIF, CGs, or,
> what I would
> > recommend, controlled English.
>
>
> JF: Even though you see no difference in principle between DB
> constraints
> and AI axioms, mightn't it still be the case that in practice
> there are
> certain kinds of axioms that AI folk typically use in an
> ontology that DB
> folk don't use in a data model? If so, could you give some examples?
MW: The constraints you get in data models are by and large restricted
to cardinality and uniqueness constraints. That is because these are
the ones that can be implemented in the structure of a database (as
opposed to its contents and operations on them).
MW: "No-one can be older than their father (barring travel at near light
speed)" is an example of a constraint data modellers would not be
concerned with.
>
> JF: If this isn't the case (i.e., if there is no difference
> in practice
> between the kinds of constraints found in DB-land and the
> kinds of axioms found
> in AI-ontology-land) then don't we need to ask something like
> "Why use an
> ontology?"
MW: I expect to see a convergence between data models and ontologies
over the next 5-10 years. One reason why I am here. There are benefits
on both sides. Data models get a greater capability, especially for
reasoning, ontologies get a bigger audience.
>
>
> Matthew responded to another email on a different thread saying:
>
> MW: I think it is just horses for courses. Data models
> grew up to support
> database design. It is not surprising that they are quite
> good if that
> is what you are doing.
>
> I'm thinking, "OK. So data models are good for database design. And
> ontologies, then, are good for ...??"
MW: Well that's my question too.
>
> One reason to design and build a database is use it to answer
> questions.
> Is the SUO group building ontologies, in part, to answer questions?
> If so, are the questions to be asked and answered of a different kind
> from those that can be put to and answered by a DBMS?
>
> Or, perhaps, because ontologies deal with different kinds of subject
> matter, the mechanisms that "ontology systems" use to answer
> questions are
> not typically found in commercial DBMS (and hence this is why we
> need an ontology system and not a database system)?
>
>
> Maybe Adam could also chime in on why he feels that the
> case study he mentioned used an ontology not a data model. I
> will study
> again the correspondence between Matthew and Adam on that,
> but I am wondering
> if Adam might be able to collect his points into a short list?
>
> Thanks,
>
> Jim
>
>