Re: new model theory for DAML+OIL

The more I look at Pat's new proposal, the more I like it.

However, there is not just a notational difference between Pat's proposal
and mine, there are real differences.


Think of it the following way:

I tried to get a model theory that was able to determine the map from
literals (05) to literal values (the integer 5 or the string "05") from
semantic information, such as range restrictions on properties.  Pat's
proposal determines this map via strictly syntactic means, i.e., from the
RDF graph.  Both proposals restrict the denotation of literals using
semantic means, so that a range restriction on age to xsd:integer would
require that all literal values that are objects of the age relationship
belong to the value space of xsd:integer (namely integers).


My proposal would have problems when local syntactic information, such as
<age xsd:integer>05</age>, clashed with semantic information, such as the
range of age is the XML Schema Datatype union of xsd:string and
xsd:integer, because it would have no mechanism for preferring the local
information over the other information.  (There might be a way of fixing
this but it would not be pretty.)

I think that Pat's proposal has some places where it might be considered to
diverge from XML Schema datatypes.  For example, if you don't have any
syntactic type information, say as in

	<John age 05>

but you do have semantic type information, say as in

	<age range XS??:(union integer string)>

then you could not deduce that John's age is the integer 5, only, perhaps,
that it is either the integer 5 or the string "05".  (Note that I'm
extending Pat's solution a bit here, at least I think that I am.)

In a strange sense, Pat's proposal is closer to the solution in DAML+OIL
than mine is.

All that said, I think that I'm willing to go with Pat's proposal.  At
least I'm willing to try to get it nailed down and see what its
consequences are.

Peter F. Patel-Schneider
Bell Labs Research

Received on Thursday, 11 October 2001 10:22:05 UTC