> I don't think you mean "the absence of a tuple". The relational model
> has the vast edifice of "null" to support absent data in a cell of a
> table, but it has never had any way of representing a missing or unknown
> row.

This depends on whether on not you are making what C.J Date calls  the
"Closed World" assumption about your relational database (which he prefers
to).  Thus the absence of a particular row in a relation is taken to imply
the falsity of the predicate that relation represents. If your employee
relation has no record for "Gary Stephenson", it is assumed to be true that
I am _not_ an employee of yours.  I think we often make this assumption
implicitly about our databases, without properly appreciating the
implications thereof for normalization etc.

> An element with empty content is surely the parallel of a cell
> containing a zero-length string in SQL. Users may use this with similar
> semantic intent to omission of the element, but it's hard to see it as
> an analog of SQL's NULL.

And may the <deity_of_choice> be praised for that!

> For my part, I have always thought that xsi:nil is an abomination.

ditto SQL NULLs (imho)

gary