RDF on the web is inherently colored

R.V.Guha wrote:

[[
Jonathan,

 We were imprecise in our language. Sorry.

 We should have said that dark triples introduce non-mon into the rest
of any language whose
semantics interprets them. DTs themselves are monotonic.

 As to your example, the issue is that the semantics of some language
(in this case OWL),
would need to incorporate a check for formulae involving the color of
terms in the formulae
wherein the color itself was expressed using other formulae. This is
what causes the non-mon.

 My intuition is that the formulae stating darkness are coming near the
realm of (but not close
enough to cause paradoxes) stating things about the semantics of the
language that we are seeing
non-mon. Any closer or any more powerful constructs and we will be there.
]]
-- http://lists.w3.org/Archives/Public/www-rdf-logic/2002May/0120.html


It seems that RDF on the web is inherently colored, and hence I don't see DT
introducing more problems that might already exists. I say this because the
"lightness" of any graph of triples is dependent on the viewing of the graph
with respect to the _current_ graph. That is, any URIrefs within the current
graph which identify _another_ RDF graph, do not take as asserted the
statements within _that_ graph.

Simplifying the situation to remove factors involving protocols and networks
(if we don't make some simplifying assumptions this gets horribly
complicated) let us assume that rather than the color of any 'ol RDF
document being "rdf:White" by assumption, instead, let us color the RDF
graph by the _base URI_ of the originating document. Now we can express the
_current document_ as a URI, and under base RDF (i.e. every triple
asserted), the set of asserted triples is the set of triples whose _color_
is equal to the _current document_ i.e. the _base URI_.

Now the process of "merging" graphs can be represented by the recoloration
of one graph either to match the second graph, or else the assignment of a
new color to both graphs (thankfully there is no limit to the number of URIs
we may use :-)

As an implementation, a 'triple' store would be able to suck in every RDF
document it pleases and store them alongside eachother (colored triples) and
yet keep track of what is asserted with respect to each document and what
isn't. This process would seem entirely monotonic to me, assuming we agree
on a unique base URI for each document ... and the web does assume this base
URI. So already, RDF seems entirely colored from a practical perspective.

Any 'RDF inferencing engine' whatever that may be _already_ has to decide
what is a truth and what isn't (or at the very least what is an assertion),
even if only to say "I accept as asserted all triples whose color is the
current base URI."

Now suppose all the triples in a 'document' aren't assigned the same color,
i.e. some aren't colored to the base URI, so what? I don't see how this
process of coloring _intrinsicly_ adds any non-monotonicity above what would
already exist. I submit that my way of viewing the RDF world is entirely
consistent and that allowing triples to be syntactically colored is just as
monotonic -- perhaps more! -- than coloring by virtue of an originating base
URI.

Jonathan

Received on Friday, 31 May 2002 08:17:35 UTC