Why contexts? (was: Klyne Contexts: 5.3-5.5 resources, languages and frames)

At 01:06 PM 11/23/00 +0100, Jonas Liljegren wrote:
>I don't understand why we must have contexts in place of resources.
>Why doesn't normal resources work?  (I hadn't read the MEMO the last
>time this was discussed.)

That's a good question, and one which I had to ask myself several times 
before I started to see why.  (Historically, it felt like the right thing 
to do, and some empirical trials with RDF seemed to bear out the intuition.)

To try and answer your question:  it has to do with constructing models for 
complex (real-world) objects and systems.  I believe that resources are too 
low-level a concept for describing complex systems.  My intuition, which I 
am planning to demonstrate through examples built over the coming months, 
is that construction of complex models requires more flexible, higher-level 
constructs than resources and statements.

I must be clear, I have no reason to believe that resources and statements 
are not _logically_ sufficient to describe complex systems -- I am not 
seeking to build a logical framework beyond that supportable by 
RDF;  rather it is the _expressivity_ that concerns me.  I am seeking to 
create structures that do for information modelling what functions and 
subroutines do for computer programming:  allow us to create higher-level 
constructs from lower level pieces.

There is one particular issue I have identified when trying to construct 
complex system descriptions:  it is difficult to construct a description 
from just nodes and arcs without having an prior knowledge of the complete 
ontological framework.  I contend that when describing complex systems we 
do not start out with complete knowledge of the ontological structure that 
will describe that system.  Development of the statements and the ontology 
move forward together.

One of my reasons for using contexts is that it gives me a framework for 
making statements that do not depend on detailed ontological 
structure.  Thus, in the 1st example you cite, I know that my car has an 
engine and a body without necessarily knowing how they are ontologically 
related to the car as a whole, and I can make meaningful statements about them.

>One example says, in part:
>
>       [MyCar] --isa--> [FordEscort]
>       [     ] --rdfc:asserts-->
>         {
>         [TheBody] ----color-----> "red"
>         [TheEngine] --capacity--> "1600cc"
>         }
>
>
>The normal way to do this would be:
>
>  S1: [my:Car] --type--> [FordEscort]
>  S2: [my:Car] --body--> [my:Body]
>  S3: [my:Body] --color--> [red]
>  S4: [my:Car] --engine--> [my:Engine]
>  S5: [my:Engine] --capacity--> [my:Capacity]
>  S6: [my:Capacity] --unit--> [cc]
>  S7: [my:Capacity] --value--> "1600"


By comparison, this form of description requires that the ontological 
relationship between the components is known.

>No need for special constructions.  Plain RDF works fine.  The "my"
>things above are placed in a local namespace.

[...]

>This is how I would do "forward" inferencing.
>
>Maby should continue this discussion in rdf-logic?
>   Haven't compared this to DAML yet...

My focus here isn't primarily on inferencing, and I'm not trying to define 
an alternative DAML or similar framework.  As far as possible, I'm trying 
to avoid issues of pure logic, other than how they affect the construction 
of descriptions.

My concern is how to use RDF in the process of constructing models of 
complex systems.  We have tried to construct descriptions of systems of 
only moderate complexity, and found that strict ontological structures tend 
to lock us into rigid and inflexible forms of expression, making the task 
of constructing a description very difficult, and definitely not practical 
for our purposes.

The end goal of this effort is to come up with a practical way to construct 
RDF descriptions of complex systems and objects.  In due course, I would 
expect this to be related to DAML-O and other ontology work in a consistent 
way, not to diverge from it.  I think here is the appropriate forum, 
because I'm not trying to define new logical structures.

Thanks for your thoughts.

#g

------------
Graham Klyne
(GK@ACM.ORG)

Received on Thursday, 23 November 2000 09:39:59 UTC