Re: What is an RDF Query?

I wonder if I might circulate part of a document which I wrote a few 
weeks ago for the RDF Core WG, since it is relevant to this debate 
about anonymous nodes. We (the WG) spent a lot of time discussing 
these issues, and in writing a draft model theory I wrote the 
following to try to help clarify the discussion. I think it is 
self-contained, even though it refers to other parts of the document. 
(The terms 'anonNode' and 'uriref' are from N-Triples.) (This is from 
a DRAFT, and will not appear in this form in a WG publication.)

The key point is that one has to be careful to distinguish between 
asserting something and querying it. Variables change their 
interpretations when this boundary is crossed. Existential variables 
in a query are treated in inference exactly like universal variables 
in an assertion (and vice versa); those are the 'real' variables that 
can be bound to new values during the inference process. Existential 
variables in an assertion act just like 'blank names' and it is 
invalid to bind them to new values.

Pat Hayes
-------

5. Publishing content: assertion versus query.

The model theory characterizes truth-preserving relations between 
expressions (graphs, or N-triple documents) but it does not specify 
what exactly is being 'said' when a graph is published. <comment> 
This is what one would expect, since publications are tantamount to 
speech acts rather than expressions. The same expression can, 
notoriously, be used to do various different things: it can be 
asserted, questioned, doubted, assented to, etc. </comment>

The most obvious assumption is that to publish an RDF graph is to 
assert that it is true, thereby in effect offering a warranty for 
anyone else to draw valid conclusions from it. Let us call this a 
descriptive or asserting publication. It can be characterized as 
making a public claim about the appropriate uses of the expression; 
in effect, it says: this can be correctly used to make inferences 
from. However, the model theory would apply just as well to a 
different kind of publication, where the intended meaning is not that 
it is appropriate to draw conclusions from the expression, but that 
inference is intended to go in the other direction, so that the 
publication says, in effect: this should be used to make inferences 
to; or, in other words, can anyone prove this? If we represent E 
entails E' by writing (E |=> E'), then assertional publication of E 
might be represented as  (E |=> ??) , putting E at the blunt end of 
the arrow, while the other case - which we can call a querying 
publication -  might be written as putting E at the sharp end of the 
entailment, (?? |=> E);  where in each case the world in general (or 
the community to which the publication is addressed) is being invited 
to fill in the ?? blanks.

This picture is almost certainly too simplistic as it stands, since 
one would presume that the intention of publishing a query is not 
merely to advertise a potential conclusion, but would include an 
implicit request to be told about any assumptions out there that 
would entail it. This could be handled by a general assumption about 
various kinds of publication acts and their associated protocols (eg, 
that if A publishes in query mode an expression containing a 
variable, and B binds a value to the variable, then B should inform A 
of the binding; what might be called a cookie-variable protocol); but 
in any case, RDF has no way to make this distinction at present. I 
mention it only to try to clarify the distinctions that have arisen 
in discussions.

We could propose a related language to RDF which might be called 
Resource Querying Format, which is identical to RDF except that RQF 
expressions are understood to be queries rather than assertions.
<joke> The syntax of RQF could be identical to that of N-triples 
except that an RQF document must end with a literal of the form 
[Canadian: ", eh?"] [US: ", right?"], etc.
</joke>
The processing appropriate to RQF would be somewhat different from 
that for RDF. In particular, anonNodes in RQF documents would have to 
be treated as genuine variables which can be bound to values at run 
time, as discussed in section 3. A unification process which binds an 
RQF variable to an RDF uriref or anonNode would be a central part of 
the machinery for linking RDF assertions to the RQF queries which 
they entail.

------

---------------------------------------------------------------------
(650)859 6569 w
(650)494 3973 h (until September)
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes