Re: Classes and predicates as first class objects

Classic had some things that Bob might have been thinking of that have some non standard
dl flavor and have some flavor of second orderness.
(for more details, the manual is obtainable from:
http://www.bell-labs.com/project/classic/lisp.html)

The semantics specifications (e.g.,
http://www.cs.washington.edu/research/jair/abstracts/borgida94a.html)
were careful to handle these topics separately however.
These all arose out of application needs.

Some thing Bob may have been remembering:

close (in everything but the initial version of classic) was a knowledge base operation
and not a constructor in the language.
The operator close took as arguments a role and an individual and then allowed one to
state that no new fillers were allowed on that role of the individual (thereby
introducing a max cardinality statement on that role).

rules - classic had 3 kinds of rules.  All took a class as an argument and some also
took filtering functions which worked on arbitrary arguments and functions that could be
used to calculate fillers.
cl-add-rule (and its variants for the more complicated forms of rules) was a function
and not a constructor in the language.

tests - classic had a notion of a test function  that could be used to test if a classic
object or a datatype satisfied a particular function.  those functions took arbitary
arguments.
This was the escape hatch for classic since it was a limited language and application
writers sometimes needed additional expressive power over what the language provided.
They could essentially arbitrarily extend the language using this feature although there
were some caveats.
Test was a language constructor and had requirements on how it was used (see page 9 in
the manual for details).


We did make claims about the usefulness of all three features in applications in
practice.  In fact, all the applications that i dealt with made heavy use of rules and
tests and heavy use of either close or max cardinality restrictions to force a closure.

deborah

Ian Horrocks wrote:

> On August 15, Bob MacGregor writes:
> >
> > This discussion started not as a question about the merits of DLs, but
> > asked what you get when you sacrifice the ability to treat classes/predicates
> > as arguments to other predicates.  Slightly paraphrasing:
> >
> > > > > Clearly, not allowing this feature (classes and arc labels as
> > > > > first class objects) buys description logics something.
> >
> > Please correct me if I'm wrong, but my recollection is that CLASSIC allowed
> > second-order syntax, and in fact made some claims as to the utility of doing
> > so.  I my recollection is correct, then its not the case that we have to make this
> > particular sacrifice to achieve the "benefits" of a DL.  Rather, the trade-off
> > exists only for certain subclasses of DLs.
>
> Classic does not have this ability - at least not in the part of the
> language for which sound and complete reasoning is claimed.
>
> Moreover, Classic is a very old system, predating most of the research
> on which modern DL systems are based and which led to the benefits
> that I have described. As I mentioned, most/all of these "modern" DLs
> are decidable subsets of FOL, and consequently none of them have the
> ability to treat classes/predicates as arguments to other predicates.
>
> Adding the ability to treat classes/predicates as arguments to other
> predicates would take DLs outside of (decidable subsets of) FOL. The
> computational properties of such languages are not well understood,
> and sound and complete (and terminating) decision procedures are not
> known.
>
> Ian
>
> >
> > - Bob
> >
> > ----- Original Message -----
> > From: "Deborah McGuinness" <dlm@KSL.Stanford.EDU>
> > To: "Ian Horrocks" <horrocks@cs.man.ac.uk>
> > Cc: "R.V.Guha" <guha@guha.com>; <www-rdf-logic@w3.org>
> > Sent: Thursday, August 15, 2002 14:31
> > Subject: Re: Classes and predicates as first class objects
> >
> >
> > >
> > > to support Ian's last statement that DLs are and have been used in a wide
> > > variety of applications and also because people had previously asked for some
> > > example applications,
> > > i include a paragraph from a message i sent a while ago on a related topic
> > > (full message available from
> > > http://lists.w3.org/Archives/Public/www-rdf-logic/2001Feb/0044.html)
> > >
> > > This was just a list of some variety of applications of an earlier description
> > > logic - CLASSIC - but it shows applications with references in broad areas.
> > >
> > > "I have helped people use CLASSIC in a series of large applications; the
> > > largest and longest lived was a family of configurators called PROSE/QUESTAR
> > > [1].  This included 17 configurators, some of which were used for a decade.
> > > They were used by AT&T and Lucent.  Other major application areas include data
> > >
> > > archeology  [2], software discovery [3], query expansion [4], query answering
> > > [5], plan representation [6],  knowledge based software engineering [7], and
> > > other domains.   We also spent time considering the usability issues of the
> > > language [8,9]."
> > >
> > > Most if not all of these applications (and the application areas in general)
> > > benefited greatly from having the qualities that Ian refers to of reliable and
> > > efficient reasoning.
> > >
> > > Deborah
> > >
> > > [1] Deborah L. McGuinness and Jon Wright. ``An Industrial Strength Description
> > >
> > > Logic-based Configurator Platform''. IEEE Intelligent Systems, Vol. 13, No. 4,
> > >
> > > July/August 1998, pp. 69-77. )
> > >
> > > [2] Ronald J. Brachman, Peter G. Selfridge, Loren G. Terveen, Boris Altman,
> > > Alex Borgida, Fern Halper, Thomas Kirk, Alan Lazar, Deborah L. McGuinness,
> > > Lori Alperin Resnick. ``Integrated Support for Data Archaeology.'' In
> > > International Journal of Intelligent and Cooperative Information Systems, 2:2
> > > 1993, pages 159--185.
> > >
> > > [3] P. Devanbu, R.J. Brachman, P.G. Selfridge, B.W. Ballard: "LaSSIE: A
> > > knowledge-based software information system" Communications of the ACM,
> > > 34(5):35--49, May 1991.
> > >
> > > [4] Deborah L. McGuinness. ``Ontological Issues for Knowledge-Enhanced
> > > Search''. In the Proceedings of Formal Ontology in Information Systems, June
> > > 1998. Also in Frontiers in Artificial Intelligence and Applications,
> > > IOS-Press, Washington, DC, 1998.
> > >
> > > [5] Alon Y. Levy, Anand Rajaraman and Joann J. Ordille, ``Query Answering
> > > Algorithms for Information Agents'' Proceedings of the 13th National
> > > Conference on Artificial Intelligence, AAAI-96, Portland, Oregon, August,
> > > 1996.
> > >
> > > [6] P. Devanbu , D. Litman , CLASP - a plan representation and classification
> > > scheme for a software information System,  Published in Artificial
> > > Intelligence , 1996
> > >
> > > [7] P. Devanbu , M. Jones , The use of description logics in KBSE systems.
> > > Published in ACM Transactions on Software Engineering and Methodology , 1997
> > >
> > > [8] Deborah L. McGuinness and Peter F. Patel-Schneider. ``Usability Issues in
> > > Knowledge Representation Systems''. In Proceedings of the Fifteenth National
> > > Conference on Artificial Intelligence, Madison, Wisconsin, July, 1998. This is
> > >
> > > an updated version of ``Usability Issues in Description Logic Systems''
> > > published in Proceedings of International Workshop on Description Logics, Gif
> > > sur Yvette, (Paris), France, September, 1997.
> > >
> > > [9] Ronald J. Brachman, Alex Borgida, Deborah L. McGuinness, and Peter F.
> > > Patel-Schneider. "Reducing" CLASSIC to Practice: Knowledge Representation
> > > Theory Meets Reality. In Artificial Intelligence 114(1-2) pages 203-237,
> > > October
> > >
> > > Ian Horrocks wrote:
> > >
> > > > On August 14, R.V.Guha writes:
> > > > >
> > > > >
> > > > >  Much of the debate around layering of OWL on top of RDF and RDFs boils
> > > > > down to whether the Semantic Web should treat classes and arc labels as
> > > > > first class objects, about which arbitrary new kinds of statements can
> > > > > be made.
> > > > >
> > > > > This is an important architectural choice which has to take into account
> > > > > results from systems that have been built. Looking at what was learnt
> > > > > from such systems would probably be productive ...
> > > > >
> > > > >  RDF, which has been largely influenced by the experimental "scruffy"
> > > > > side of AI has gone the route of many experimental AI systems (starting
> > > > > from KRL, RLL, .... CycL) and incorporated these as first class objects.
> > > > > In my experience, and the experience of the builders of these systems,
> > > > > this has been a useful feature. Description Logics, which come from the
> > > > > more "neat" side of AI chose not to allow this ...
> > > > >
> > > > > Clearly, not allowing this feature buys description logics something.
> > > > > Ian, maybe you could explain exactly what this is and how it has been
> > > > > found useful in large DL systems that have been built?
> > > >
> > > > Work on DLs has resulted in the development of a family of logical
> > > > languages with precisely defined semantics and well understood
> > > > computational properties. They are (almost invariably) decidable
> > > > subsets of FOL and are closely related to propositional modal and
> > > > dynamic logics.  For many of these languages, provably sound and
> > > > complete decision procedures have been devised. Several DL systems
> > > > have been based on optimised implementations of these algorithms, thus
> > > > providing users with reasoning services that are both reliable and
> > > > efficient. These systems are being used in a wide range of
> > > > applications, e.g., in medical-informatics, bio-informatics, chemical
> > > > engineering and geographical information systems.
> > > >
> > > > Ian
> > > >
> > > > >
> > > > > thanks,
> > > > >
> > > > > guha
> > > > >
> > > > >
> > > > >
> > > > >
> > >
> > > --
> > >  Deborah L. McGuinness
> > >  Knowledge Systems Laboratory
> > >  Gates Computer Science Building, 2A Room 241
> > >  Stanford University, Stanford, CA 94305-9020
> > >  email: dlm@ksl.stanford.edu
> > >  URL: http://ksl.stanford.edu/people/dlm/index.html
> > >  (voice) 650 723 9770    (stanford fax) 650 725 5850   (computer fax)  801 705
> > > 0941
> > >
> > >

--
 Deborah L. McGuinness
 Knowledge Systems Laboratory
 Gates Computer Science Building, 2A Room 241
 Stanford University, Stanford, CA 94305-9020
 email: dlm@ksl.stanford.edu
 URL: http://ksl.stanford.edu/people/dlm/index.html
 (voice) 650 723 9770    (stanford fax) 650 725 5850   (computer fax)  801 705 0941