RE: requirement: rdfs query (for lack of a better name...)

> Where in my requirement does it say "just a little bit of inference"?
> Where does it imply it? I must have missed that bit. :>
> 
> And, besides, charter thwack:
> 
>   1.8 Derived Graphs
> 
>   The working group must recognize that RDF graphs are often
>   constructed by aggregation from multiple sources and through logical
>   inference, and that sometimes the graphs are never
>   materialized. Such graphs may be arbitrarily large or infinite.

We've been through this before, and I'm getting a bit tired of it.
This sentence seems to be pointing out exactly what I've been saying: we
can't standardize an implementation because implementations will have
very different requirements depending on how they get their knowledge
about the underlying data model.
This paragraph even seems to make explicit that we're meant to be
addressing how we query a graph, not some of the particular ways to form
that graph from.
If you've got some RDF and some RDFS, then there are two quite
reasonable graphs: the raw RDF the way it's written, and a "completed"
graph which has a whole bunch of inferred triples. (There are also a
number of other graphs which can be derived from this data set in other
ways, such as the possible graphs, the impossible graphs, the graphs
with the directions of all edges reversed, the interesections or unions
of any of these other graphs, etc.)
An implementation should have the freedom to interpret a query against
any of these "virtual" graphs, thus we cannot specify how an
implementation operates.

> > What's more, adding support for just one particular flavor of
> > supplementary semantic knowledge (RDFS) is great way to 
> kill off use of
> > any other knowledge sources. 
> 
> Uh... RDFS is hardly "one particular flavor of supplementary semantic
> knowledge". That's about as perverse a description as I can
> imagine. Besides, if you're worried about "killing off" OWL, that's an
> argument to do RDFS in 1.0 and OWL in 2.0, not an argument not to do
> RDFS at all.

I think it's quite clear that sucking every possible type of querying
into our specification is about the worst way to write a standard
possible.
The fact that inferencing is very useful is an excellent reason to make
sure our query language is compatible with use where RDF is the data
model on which inferencing occurs.
The fact the inferencing is a complex and open-ended problems and that
we could not possibly address it (because the basic semantic encodings
are still being standardized) is a pretty incontrovertable argument that
we can't solve the inferencing problem. I think that's a good reason not
to try.

Let's leave derived graphs and inferencing to working groups that know
how to address them and confine ourselves to representing queries
against the data model which underlies it all.

Received on Thursday, 6 May 2004 19:05:58 UTC