new semantics initiative

Greetings. This is a heads-up message to let y'all know what Guha and 
I have been up to this last week or so, and to ask for the WGs 
co-operation. We should be able to give you all a much more complete 
version in time for the Bristol F2F, including draft documents in a 
couple of days.

Currently, each web language (RDF, RDFS, DAML+OIL, OWL,...) has its 
own semantics and its own syntax. The various groups have succeeded 
in mapping all the various syntaxes into RDF triples, because this 
requirement was chartered; but the various model theories have not 
been aligned so successfully because, frankly, that is more difficult 
to do, and unexpected problems of 'layering' have arisen and have 
proved very hard to resolve. However, this situation seems peculiar 
when looked at from a logical perspective, since these various 
languages are all pretty much sub-languages of a very well-understood 
'textbook' logic with a very simple and  uncontroversial semantics, 
which is classical first-order logic. So Guha has proposed that we 
make this intuitive mapping into a part of the formal specifications, 
in an effort to produce some overall coherence to the various web 
languages in use now and in the future.  (This idea is based on 
earlier work in AI in the 1980s, when an explosion of alternative 
knowledge representation formalisms were being invented, and in spite 
of their various pragmatic or philosophical goals, the relationships 
between them were clarified by mapping them all into first-order 
logic. Even those aspects which did not map into FOL - and there were 
some, particularly to do with nonmonotonicity - were clarified by the 
exercise, in fact. Guha and I were both influential in that effort: 
me by mapping frame languages and 'procedural' Krep languages into 
FOL, he by mapping CYC into FOL.)

So, the proposal is to write two new documents and to slightly modify 
the RDF MT document. The modifications to the MT will amount to 
little more than a revised introduction which points out that the 
other two documents, taken together, amount to an alternative way of 
describing RDF semantics which is entirely formally equivalent to 
this one, but may have certain advantages for some developers. (There 
is also a related proposal to allow unasserted triples, in a way that 
overcomes the nonmonotonicity issue that the 'owl:Dark' proposal 
raised. Details of that will follow in later messages.)

The other two documents are (1) a description of the overall 
technique, including a description of the semantic language Lbase, a 
model theory for it and a general discussion of how to use it to 
provide semantics for other languages) , and (2) a document using 
this technique to give a semantics for RDF, and relating that to the 
RDF MT, proving that they are equivalent. Another way to think about 
this second document is that it provides an alternative way of 
describing the RDF MT which will be accessible to anyone with a 
background in logic, and which will also provide machine-checkable 
renderings of RDF meaning in a general-purpose framework which allows 
RDF formal content to be connected to, for example, DAML+OIL and OWL 
formal content in a uniform framework.

Guha and I are writing the first of these in any case, but we believe 
that it would be very useful and beneficial in the long run if the 
'official' account of RDF also incorporated the second document, 
which we should have drafted in a few days. I would like therefore to 
place this on the agenda for the F2F. If there are any questions that 
people want answered before the F2F, feel free to email.

Since the Lbase RDF(S) semantics and the RDF(S) MT align exactly (you 
get the same interpretations and the same entailments either way) it 
would be possible to merge the two documents into a single document 
which gave both versions of RDF semantics in parallel. This was 
Guha's original idea; but after trying to do it, my own preference is 
to have two documents, for several reasons. At this stage it is 
rather traumatic to make such a major change to the MT document, and 
would probably take longer than writing a new document from scratch; 
the parallel development is likely to be confusing to readers; the 
Lbase semantics requires one to understand a more complex model 
theory than the RDF MT alone does; and the two-document presentation 
allows the MT document to continue to be phrased as a kind of 
intro-to-MT for non-logicians, while the RDF-in-Lbase document can be 
written more tersely for readers who are already logic-savvy. (It 
will also have much shorter proofs of the lemmas, by the way, since 
it can appeal to established results like Herbrand's theorem and 
compactness.) I hope the WG approves of this decision, which (in 
spite of the extra document in the spec) I think provides the widest 
utility to the largest number of readers and at the same time 
provides a firm link between the RDF spec (which once the WG has 
finished with it will be cast in stone) and what will, we hope, 
become the standard way to relate content between different semweb 
languages.

We hope to have a draft version of both documents (largely assembled 
from cut/paste of existing text) by early in the week.

I apologize that the re-drafting of the MT document has been delayed 
by this. The likeliest prediction is now Tuesday. Since the changes 
will all be intro. text, minor editorials and bug-fixing, I hope this 
will not pose any severe problems for the F2F.

Pat Hayes

-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes

Received on Monday, 10 June 2002 13:09:58 UTC