Thanks a lot for the answer. I can see the problem now. Actually you may
want to consider an alternative in the official PREMIS schema that
requires less maintainence:

1. Define an abstract objectType, with neither type attribute nor
objectCategory element defined at this level

2. Define a global objectCategorySimpleType which is basically the
xs:string type, like the following:

<xs:simpleType name="objectCategorySimpleType">
  <xs:restriction base="xs:string"></xs:restriction>
</xs:simpleType>

3. Define three or even more concrete objectType that extend the
abstract objectType, in which they all have objectCategory elements
which are defined respectively, although of the above defined
objectCategorySimpleType, which is basically the xs:string type, but fix
the value to the controlled vocabulary. The following give the
definition of the fileObjectComplexType as an example:

<xs:complexType name="fileObjectComplexType">
  <xs:complexContent>
    <xs:extension base="objectComplexType">
      <xs:sequence>
        ...
        <xs:element minOccurs="1" maxOccurs="1"
          name="objectCategory" type="objectCategorySimpleType"
          fixed="file">
        </xs:element>
        ...
      </xs:sequence>
    </xs:extension>
  </xs:complexContent>
</xs:complexType>

I think this kind of schema structure has a handful of benefits to offer
to the PREMIS community:

First, it eliminates possible confusions introduced in the following two
ways:

1. An object with a "file" type attribute value, but a different
objectCategory value
2. An object with the same type attribute and the objectCategory values,
e.g., "representation", but includes an element that is not allowed in
that type of object, e.g., objectCharacteristics

Although common sense should not allow the above happen, in practice you
never know. Errors like this happen, either from human or from the
machines. If the schema does not capture them and allow them validated,
chances are the errors will be introduced into the system.

Second, not only this schema follows the objectCategory definition as an
xs:string, but it strictly limits its value when comes to the already
defined three types of object, file, bitstream, and representation. On
other hand, if any implementation needs more flexibility, the structure
of the extension is already provided. People only need to extend their
own concrete objectType, mostly by making use of the many globally
defined types to construct the extension. For any extensions following
this example, the extended objectCategory should have the "fixed"
attribute that limit the value to a controlled vocabulary.

Hope this helps,

Zhiwu Xie

Graduate Research Assistant
Research Library
Los Alamos National Lab


On Thu, 2006-02-09 at 10:38, Jerome McDonough wrote:
> > On Mon, 6 Feb 2006, Zhiwu Xie wrote:
> >
> >> This is an implementation question.
> >>
> >> In schema v1.1 there's an attribute "type" for the "object" element.
> >> Although optional, it's set to enumerate either "file",
> >> "representation", or "bitstream". I think this attribute is redundant
> >> because the objectCatgory contains the same information.
> >>
>
> Well, yes and no. Yes, in that they're intended to convey the same
> *type*
> of information, but no, in that objectCategory doesn't insist on the
> controlled
> vocabulary established in the PREMIS model, so in practice, someone
> could put any type of information they wanted in there. My memory of
> the discussions was that the flexibility in objectCategory was
> intentional, as
> people were a bit reluctant to state that everyone in the world had to
> conform to the PREMIS' group's notion of object categories. So,
> objectCategory
> was left as an open string, in case someone decided they needed
> flexibilty
> in local practice. The type attribute is obviously more constrained,
> and was
> included so that there could be a dependable, controlled vocabulary for
> use between institutions for identifying the category of object.
>
> >> Also it hinders further extension, because in case of more object
> >> categories are needed, this attribute needs to be modified each
> >> time an
> >> extension is proposed.
> >>
>
> That is true, and that is the downside of controlled vocabularies.
> They do
> require maintenance. The upside is having a reliable shared vocabulary
> between institutions/applications. So, there's a cost/benefit
> calculation in
> deciding whether to adopt them. Part of the calculation is the
> likelihood
> of rapid change. To date, I don't think I've heard anyone say that they
> need the set of object categories altered or extended, so I don't
> think it's
> imposing a particularly great burden to date, particularly given that
> the
> type attribute's use is optional.
>
> >> Even no extension is expected, it still adds unwanted complexity to a
> >> hierarchical schema.
> >> I'm working on a schema that first defines an
> >> abstract objectType, then extend it to three or even more different
> >> types, e.g., fileObjectType, bitstreamObjectType,
> >> representationObjectType. If we must have this attribute, I'll
> >> have to
> >> further restrict the extended types to accommodate this redundant
> >> attribute.
> >>
>
> I'm afraid I'm not sure I see your problem. If you're developing
> your own local application profile schema to support PREMIS,
> your only real concern is to support the *mandatory* aspects
> of the PREMIS schema, and the type attribute isn't mandatory.
> So, what exactly is preventing you from creating an abstract
> objectType schema which omits the "type" attribute and extends
> it from there?
>
>
> Jerome McDonough, Asst. Professor
> Graduate School of Library & Information Science
> University of Illinois, Urbana-Champaign
> 501 E. Daniel Street, Room 202
> Champaign, IL 61820
> (217) 244-5916
> [log in to unmask]