Re: What do the ontologists want?

> > From: pat hayes [mailto:phayes@ai.uwf.edu]
> > [Jonathan Borden wrote:]
> > > Certainly being able to quote statements/triples is useful 
> > > ... indeed a practical requirement.
> > 
> > Can you (or anyone) say why the ability to quote is considered a 
> > practical necessity? From where I am standing it seems an arcane and 
> > exotic ability, not one that is of central practical importance. What 
> > is the practical utility of being able to refer to a predicate, 
> > rather than use it?
> 
> I've got sympathies on both sides of this, based on past and present systems
> I've used; summarised below.  I'm also pretty hard-nosed about why I'm doing
> this; summarised in the last paragraph.
> 
> Some systems are more obvious to design if there is a facility to make a
> statement about a statement.  Almost always, these are attributions: the
> typical example seen on this list, and the examples we used in SMK, are of
> the form "X says 'statement Y'".  (Question: Can anyone come up with a
> different use for reification?  If this is the only special-case, should
> there be a different mechanism for attribution?)

?  The RDF spec itself has a different use case.  Furthermore, all the uses 
are pretty old hat (at least on www-rdf-interest).  I'm sure I'm mising 
something, but just in case, general categories include

Attribution: your "Midwinter Spring is its own season", says Eliot

Quantification: Temperatures are in the high 30s (in degrees centigrade)

Qualification: "The check's in the mail", he lied (OK, it's 
attribution+qualification, but I like the example)

Examples that fall into the above bin include

Temporal placement:  1980: The capital of Nigeria is Lagos; 1990: The capital 
of Nigeria is Abuja

Confidence factors: There is a 15% likelihood that Malaria results in death.

> We were working with clinical systems, where the ability to attribute
> particular statements to clinicians was of legal importance in case of
> lawsuits.  No attribution, no adoption of system.  We chose to implement
> this using reified triples; this opened up a can of worms as far as the
> implementation was concerned, as we could then make statements about the
> statements of attribution, and we could also construct self-referencing
> systems such as:
> 
> 		#2 writtenBy peter (triple ID #1)
> 		#1 writtenBy peter (triple ID #2)
> 
> Overall, the approach caused more problems than it solved; GRAIL and the
> more recent work at the University of Manchester dropped this facility, and
> we put in special-case code for dealing with attribution.

I think the answer to these problems are effective tools.  Rules-based systems 
should be able to operate on the metadata making up attributions to set 
sensible limits ("sensible" being determined by the field of use).

> However, we were dealing with a knowledge base that was totally under our
> control.  One of the interesting features of RDF is that it allows an author
> to mark up a source over which they have no direct control; one view of
> search engines and classifications like Yahoo! is that they will evolve into
> huge metadata repositories rather than simply free-text engines.  In this
> situation, there are a couple of areas that need thought:
> 
> 1) How does a third party refer to portions of another's work?  Especially
> if that work is RDF?  For example, I might want to say that I agree with all
> the statements in a particular document except statements A, B and C.

Well, while this is rather cumbersome in RDF, it's quite possible.  You can 
either list all the statements you agree with, or provide an exception 
predicate which overrides a statement of agreement with the whole document.

> 2) (a) Should one be able to reason about such statements / does it gain us
> anything; (b) How should one be able to reason about such statements if it
> does gain us anything?
> 
> Apart from these areas, I agree with Pat: it is an arcane and exotic
> ability.  It is also an ability that will cost us very dearly in terms of
> being able to reason about the resulting (rather baroque) structures.

Hmm.  I understand how reification can complicate matters, but I don't see it 
as being more insurmountable than the complexity of distributed ontology in 
the first place.


-- 
Uche Ogbuji                               Principal Consultant
uche.ogbuji@fourthought.com               +1 303 583 9900 x 101
Fourthought, Inc.                         http://Fourthought.com 
4735 East Walnut St, Ste. C, Boulder, CO 80301-2537, USA
Software-engineering, knowledge-management, XML, CORBA, Linux, Python

Received on Tuesday, 15 May 2001 12:08:06 UTC