XML Schema Types - xsd

Question

If I have an element, animal:
<animal name="dog"/>
Which can take the following as values to the name attribute:
dog
cat
bird
$(ANY_STRING)
Where $(ANY_STRING) is a simple name, value substitution that some software will perform and validate later (ANY_STRING being literally any string). What would the XML Schema for this element look like? Restricting on the three known names is easy enough:
<xs:simpleType name="AnimalNames">
<xs:restriction base="xs:string">
<xs:enumeration value="dog"/>
<xs:enumeration value="cat"/>
<xs:enumeration value="bird"/>
</xs:restriction>
</xs:simpleType>
Restricting on $(ANY_STRING) is similarly easy on its own (using xs:pattern to restrict). But since attributes may only be simple types, is it possible to specify that the attribute may be in the list of enumerations or the $(ANY_STRING) value?
Another option I've considered is restricting on the below pattern:
<xs:pattern value="dog|cat|bird|$(.*)"/>
Although that gets pretty nasty as the list of possible values grows.
Of course, the simplest option is to just declare a string type, but I'd like to be more restrictive than that.

Answer 1

One way to define semi-closed lists like this, with a set of well-known values explicitly specified for documentation, and then with an escape-hatch to allow other strings as well, is to define your attribute with a union of your AnimalNames type and xs:NMTOKEN or xs:Name (or whatever built-in or user-defined type best captures your constraints on the other names not enumerated).
As guidot points out in his comment, such a union accepts the same set of values as its most inclusive member type, so for schemas whose sole purpose is gatekeeping it's pointless. The technique is useful for documentation and for type-driven dispatch (if the validating member type is AnimalNames do X else do Y).
In order to accommodate schema validators that don't provide information about which member type validated a particular value of a union type, some vocabulary designers do as guidot suggests, and provide two element types, one for known / predicted / expected animal names and one for other names. Or they specify the type AnimalNames as accepting the strings dog, cat, bird, and other, define the attribute name as having type AnimalNames, and define another attribute (call it other-name) which is defined as having meaning if and only if name="other". So dogs are described using
<animal name="dog"/>
and hamsters with
<animal name="other" other-name="hamster"/>
That makes it fairly simple to handle the well known names specially while still accepting other names.