Hi Muriel,

> Jumping into the conversation, it seems to me that if the
> collection description is based on ISIL identifiers, it means
> that the collection is defined by an institution.

I think Juha's initial proposal was that the collection identifier
should be based on an ISIL and that that should act as  metadata
signalling owner of the collection, but there were some arguments made
against that (on the basis that ownership changes and therefore the
persistence of the identifier becomes unreliable)....

> I imagine
> the case of a description of all French monuments by an
> American university for example because of a specific project
> on world monuments, it creates an identifier for this based
> on the US university, then somebody in France in charge of
> those monuments might feel weird about using an identifier
> based on a US university.

... I think you are going beyond the initial argument against using
ISILs. I think you are arguing that it's not the fact that ownership
changes that is problematic, but rather that the potential user of an
identifier might - because of social, political, cultural factors -
resist using an identifier, not on the grounds that it doesn't serve as
a useful identifier, or that it carries inaccurate metadata, but purely
because it was assigned by an agency that they have some social,
political, cultural "opposition" to?

I suspect you may be right! ;-) But I'm really not sure we should rule
out the use of an identifier scheme on these grounds? There is nothing
to stop a French institution assigning a new identifier which they
prefer - the perceived (social, political, cultural) benefits would have
to be weighed against the costs involved in assigning/managing a new
identifier, not just to the assigning agency but to users who have to
manage the fact that these two identifiers refer to the same collection.

> This might not be the best example
> but it seems to me that defining a collection according to an
> institution might lead to several problems.... this could
> also be the case with a set of Websites or whatever which are
> not physically in a single location.
>
> All our information systems are based on items and the
> definition of a collection are "a set of items" or
> aggregation of items if I remember well in Andy Powell and
> Michael Heaney documents. As Sarah pointed out, when dealing
> with digital items especially, collections are combinations,
> recombinations and sub-combinations of items so that ideally
> the collection identifier is the sum of all item identifiers.

I'm afraid I didn't understand this bit. If the collection identifier is
somehow derived from the identifiers of the items in the collection,
doesn't that mean that when item 101 is added to my collection of 100
items, the identifier of my collection changes? I don't think that is
what we want? Yes, in _some_ cases, as Gordon suggested, we may indeed
want to identify every possible permutation of items as a distinct
collection, but it seems to me certainly in some - if not most - cases
we want to continue to refer to the collection as the same resource,
even as it accrues new items.

> We only need, as Terry said a collection level description
> identifier (OAI provides something like this or just
> OpenURL-like sum of descriptive elements...?).

Leaving aside the OpenURL-like case for a second, I must admit I don't
really understand the suggestion that assigning/managing/disclosing
globally unique persistent identifiers for the collection-level
_descriptions_ is easier than assigning/managing/disclosing them for the
collections! Sorry, I think I'm really missing something!

Yes, I think it may be a good thing to provide unique identifiers for
CLDs too. And I'm happy in principle that I can refer to a collection
using something like:

the collection which is described by CLD-1234

But that means I need a (persistently globally uniquely identified)
collection-level description for every collection I want to refer to. I
just don't really see that that gets us much further forward! We've just
shifted the problem to identifying the CLD, haven't we?

Turning to the OpenURL-like case, (and I'm definitely not an expert at
all, so please correct me!)

As far as I can see - and I may be wrong - the attributes we are
recording about a collection are not sufficient to support an
unambiguous By-Value Metadata Descriptor. As we've already argued re
owner - some of the attributes we would explicitly exclude because their
value is liable to change while the identity of the collection does not.
Maybe this is a failing of the DC CD AP that we need to address? But at
the moment I can't see how this can work.

(Maybe we could construct a By-Reference Metadata Descriptor but as
above that just shifts the problem to identifying the CLD?)

> I imagine that a metasearch service would be able to manage
> the collection description and associate the list of item
> identifiers included in the collection. That allows to
> perform collection level dedupping, provided that the item
> identifiers are updated.
>
> The only problem is then to associate the item identifiers to
> the collection systematically....

But I think you're thinking of particular classes of collection and one
particular application area, though? It seems to me we need to be able
to describe, identify, and refer to collections for which no item-level
descriptions exist or will ever exist. So we can't rely on having a set
of item identifiers to help work out that
the-collection-described-by-this-CLD is the same as
the-collection-described-by-that-CLD.

Pete