Re: re classification scheme

[Terry Allen <tallen@sonic.net>]

Jon wrote in response to Len:
| | Several of us at NIST have been playing around with different ways
| | to define and exchange classification schemes for the OASIS
| | Registry/Repository.
| | [...]
| | We'd be interested in any comments -- in the absence of comments
| | this ER-diagram and the ClassificationScheme DTD will appear in
| | the next candidate Registry/Repository specification.
| 
| This seems to be saying that whatever "several of us at NIST"
| decide is by default what's going to be in the next specification.
| Is this really your position?

I believe the "candidate specification" is an effort by the NIST
and XML.org implementors to find common ground quickly; it isn't
a committee document.

| | 1) Every classification scheme is registered as a Registry
| | Item. Thus it has a global uniform resource name (URN) assigned by
| | the registration authority. That name can be used to access all of
| | the scheme's metadata, e.g. aliases, descriptions, dependencies,
| | etc.
| 
| This seems to be saying that the repository can only use
| classification schemes that are themselves resident in a
| repository organized in the same way as our repository.  That cuts
| out a lot of classification schemes, doesn't it?

I belive the point is that a classification scheme that isn't
accessible to a machine isn't useful (I agree entirely), and
that if it's marked up to be machine-accessible it should be
registered so you can know what version you're using, and so on
(seems like a good idea to me).  But I haven't yet
reviewed Len's model.

| Later on you say:
| 
| | Recall that a registry item cannot be classified according to a
| | classification scheme unless that scheme is itself registered.
| 
| I wasn't paying attention when this principle was adopted, but it
| seems to render pointless a lot of discussions we've been having
| recently about how to classify schemas.  No need to worry about
| choosing among registered classification schemes if there aren't
| any.

>From the outset we've assumed (because 11179 does) that c.s.'s 
would be registered items.  The present XML.org proposal, I believe,
is to use a c.s. that isn't even machine-accessible; that won't
allow many useful kinds of processing of the metdata. 

| | 2) We need a DTD to represent the classification scheme itself. An
| | OASIS classification scheme representation must validate to this
| | DTD.
| 
| Why do classification schemes need DTDs?

So they can be marked up and made machine-accessible.

| Take NAICS, for example.  Are you saying that we have to come up
| with a DTD for the NAICS before it can be used as a classification
| scheme?  I can find out about the NAICS system by looking at their
| web site.  How is my ability to correctly categorize an industry
| under one or more NAICS codes dependent on the existence of a DTD
| for NAICS?

What's the point of using NAICS in an XML-based registry if you
have no machine-accessible representation of NAICS?  Sure, you can
provide an NAICS code, but how can I process this information
in the context of NAICS's structure *by machine*?  You have two
data structures that aren't attached to each other.  Far better
to set them both up to allow linking, implicit or explicit.

I'm pretty sure the EBXMLlers (different context, to be sure)
would insist on a machine-accessible representation for EVERY
c.s., and on having them registered.


regards, Terry