There are some restrictions on supported PML schemas and naming conventions
used:

1. PMLREF pointers can be converted to node relations. For this
   to work, several conditions must be met:

   a) pointers must refer to IDs that are unique over the whole data
      set (corpus)

   b) pointers must refer to IDs of objects with the role node

   c) one must specify a mapping like in the following example:

   Assume we have two layers, A and B and assume A-layer nodes
   refer to B-layer nodes (of types B1 and B2) like this:

     <type name="A1.type">
       <structure>
         ...
         <member name="b1"><cdata format="PMLREF"/></member> <!-- pointer to B1 -->
         <member name="b2"><cdata format="PMLREF"/></member> <!-- pointer to B2 -->
       </structure>
     </type>

   To be able to represent theses pointers as PML-TQ relations, we have
   to specify the following mapping for layer A:

     A1/b1=B:B1
     A1/b2=B:B2

   which indicates that the type path (not attribute path) !A1.type/b2
   in the A-layer schema points to nodes of type B1.type in the
   B-layer schema.

   PML2Base convertor can only convert PMLREF data to relations if in
   the schema each occurrence of <cdata format="PMLREF"> corresponds
   to exactly one target-node type. This means that in the above example,
   we cannot rewrite our A schema like this:

   <!-- layer A schema -->

   <type name="A1.type">
     <structure>
       ...
       <member name="b1" type="pmlref.type"/> <!-- pointer to B1 -->
       <member name="b2" type="pmlref.type"/> <!-- pointer to B2 -->
     </structure>
   </type>
   ...
   <type name="pmlref.type">
     <cdata format="PMLREF"/>
   </type>

   since there is only one occurrence of PMLREF type. However, we can
   rewrite it as follows:

   <type name="A1.type">
     <structure>
       ...
       <member name="b1" type="pmlref-b1.type"/> <!-- pointer to B1 -->
       <member name="b2" type="pmlref-b2.type"/> <!-- pointer to B2 -->
     </structure>
   </type>
   ...
   <type name="pmlref-b1.type">
     <cdata format="PMLREF"/>
   </type>
   <type name="pmlref-b2.type">
     <cdata format="PMLREF"/>
   </type>

   and specify the mapping for layer A like this:

     pmlref-b1=B:B1
     pmlref-b2=B:B2

2. type names in PML schemas should be of the form "<somename>.type"
where <somename> is a name and ".type" is literal. The ".type" suffix
will be dropped when the type is presented to the user.

3. Different type names must have unique names across all schemas used
by the treebank; the root element name must be different for all
layers in the treebank.