Summary and some analysis: New Semantics Initiative

Here is my summary of the new semantics initiative as I understand it so far:

There are two separable issues being discussed under this title:

   o using a melange of model theories to define the semantics of SW 
languages isn't going to work, as demonstrated by webont

   o a mechanism for darkening some triples in a non-monotonic way.

I muddied the waters by introducing the second into this thread.  That was 
a mistake.  This summary is just about the first issue.

It is proposed to produce a W3C note which outlines a strategy for defining 
the semantics of SW languages in terms of Lbase, a representation of FOL.

A new deliverable by RDFCore WG is proposed.  This will define the 
semantics of RDF in terms of Lbase.  It is proposed that this document is 
in addition to the model theory document and will define exactly equivalent 
semantics to the model theory.

This proposal is motivated by a concern that:

   o model theory is hard:  it will take a lot (more than available) of 
expert effort for each SW language to define its semantics as a model theory

   o it is not clear how the model theories of multiple SW languages can be 
combined to provide a clear semantics for mixed language documents

The proposed new document is expected to have minimal effects on the 
content of the RDF/XML syntax doc, the primer, the RDF schema doc and the 
test cases doc and only minor effects on the model theory doc.

It is suggested that it is important to include this new deliverable in the 
current round of deliverables by RDFCore so that it has normative status 
enabling other WG's (webont?) to build on it. [I'm not sure about this one 
- see below]

It is suggested that this new form of representing the semantics will be 
easier for folks to understand.

Jos has tested the new proposal in Euler and found no bugs.

It is suggested that this is a harmless change; if it isn't adopted by say 
webont, it will still do some good and no harm.

I have spoken to some of the webont folks here, (Horrocks, PFPS, van 
Harmlen) and got a mixed response.  One criticism is timing; a concern that 
layering is a deep and hard issue that needs careful consideration, not 
rushing in at the last minute.

I assume that the proposed new document is not normative as it is based on 
Lbase which is defined in a note.  Is this a problem?  At least it ducks 
the question of which is right if the axiomatic and model theory semantics 
disagree.

I have a number of process concerns about this, in particular I would like 
to ensure that the whole community expert in this issue is involved working 
it out, but that doesn't sound like a good thing to do to RDFCore right now.

Brian

Received on Wednesday, 12 June 2002 13:37:20 UTC