There's an old Internet saying:  "when the only tool you have is a 
chainsaw, everything starts to look like a tree".  I'm wondering if my 
musings on contexts are in danger of falling into this trap.  No mind, the 
problems I am trying to tackle are real enough, and if the best solution is 
something else then that's fine by me.

I'm trying to capture in RDF the following kinds of idea, currently using 
contexts (and, in particular, the idea of a context-as-language, with its 
own binding of names-to-things).

Example:  I want to describe "My car", by reference to a make and model, 
but with local option selections.  Using a context to contain a collection 
of statements that describe my car:

[MyCar] --isa--> [FordEscort]
[     ] --ist-->
   [TheBody] ----colour----> "Red"
   [TheEngine] --capacity--> "1600"

I am using here the notation "[<context>] --ist--> {<StatementSet>}" to 
capture the idea of a collection of statements whose reifications are 
asserted to be true in the subject context.

But there is far more to know about my car than its colour and engine 
size.  More than I could possibly know, so I want to use a reference to a 
description (presumably from Ford) that can contain all this 
information.  This is represented by the link "--isa--> [FordEscort]".  The 
resource [FordEscort]  is, in turn, a context containing a collection of 
statements about this kind of car:

[FordEscort] --ist-->
   [TheBody] ----style-----> "Hatchback"
   [TheEngine] --fuelType--> "Petrol"

My problem is this:  when [FordEscort] refers to [TheBody] and [TheEngine] 
in reference to my car, how can I infer that these statements refer to 
parts of my car rather than some other person's Ford Escort car?  I had 
been thinking to tackle this using the association of language 
(specifically, binding of names to things) with a context, the idea of 
referencing one context from another creating a nesting of contexts (along 
the lines of "ENTER Context1" then "ENTER Context2" from the McCarthy/Guha 
descriptions), and with some lifting rules allowing names to be scoped from 
a referencing context to a referenced context.

This seems to be at odds with using URIs to identify specific things.  But 
it seems to have some resonance with Brian's and DanB's discussion of 
anonymous resources.

Finally, I have wondered if I may be going over well-trodden ground of 
frames and ontology work here.  If this has already been tackled for RDF, 
I'd appreciate some pointers to relevant work.


Graham Klyne

Received on Wednesday, 18 October 2000 10:49:53 UTC