Integration language = 
      glue language + data source description language


In the Rodin project at Wisconsin --- in which we've thus far written
more code than papers, so don't ask me for a nice overview :-) ---
we've been using Coral as, effectively, the integration language.  Our
experience makes me feel optimistic about the use of such a language
for integration; I think Jeff's proposal of a Datalog like language
will work.  (I'm not sure whether Prolog or SQL syntax will ``fly'',
but there's much to be said for Prolog syntax in this context, notably
that it lends itself well to explanations.  When your answers depend
on someone else's data, explanations tend to be useful.)

However, the ability to deal with set, list and other types of data
cannot be avoided, I think.  In this regard, I agree with Peter's push
for a richer type facility.

My preference would be for a Datalog-style language in which field
values are drawn from a suitably rich set of types, perhaps even
with a mechanism to define new ADTS --- although this last point will open a 
big new set of issues, and we should probably punt on this for now.

While this would constitute an adequate language for integration
purposes (for `gluing' together multiple pieces of information), integration
decisions will depend upon *descriptions* of what a data source
has to offer.  One can take the position that the wrapper around the
data source will accept queries in the core language, but:
(1) Not all sources one is interested in will have such wrappers.
(2) Independently of this point, we need to know what the answers mean.

The description language, I think, needs to be driven by a careful
consideration of the range of sources out there that we want to be
able to integrate.  This is where I see many of Rick's concerns pointing
to.  (And these considerations will also, I suspect compel the glue
language to provide some kind of open-ended hook like an ADT defn 
facility or an external call mechanism or ...)

Summary: I think there are really two components to an integration
language --- the `glue' language used to gather together information
from a variety of sources into a convenient user-level view, and the
`description' language used to describe what a data source has to
offer.  Mediators/wrappers need to be aware of both.