ED A35:  Overview of TEI Main DTD Files
         C. M. Sperberg-McQueen, 6 March 1993
 
We have the following main DTD files (or classes of files); 'main' DTDs
are used for texts, and are distinguished from 'auxiliary' DTDs, which
are used for meta-information (writing system, feature system, structured
header, tag set).
 
    tei2.dtd - main DTD file, used by all TEI documents
    teigis2.ent - defines parameter entities for TEI GIs
    teiclas2.ent - defines element classes and content models
    teiXX2.ent - defines parameter entities for one tag set XX
                 (so:  teipros2.ent, teidict2.ent, teilink2.ent ...)
    teiXX2.dtd - defines elements and attlists for one tag set XX
                 (so:  teipros2.dtd, teidict2.dtd, teilink2.dtd ...)
 
TEI2.DTD has the following structure:
 
  I.  Preliminaries
      declare TEI keywords (now:  %INHERITED, %ISOdate)
      embed %TEI.elementNames (= system 'teigis2.ent'), which
         declares all the <!ENTITY % n.p 'p'> entities
      declare all tag-set selection entities (TEI.prose, TEI.verse, ...)
         as 'IGNORE'
  II.  Define element classes
      embed %TEI.elementClasses (= system 'teiclas2.dtd'), on which
         see below
  III.  Define TEI.2, TEI.corpus.2 (or TEI.2.corpus, or whatever it
         is now named)
  IV.  Embed local extensions, core tag sets, base tag set, and
         additional tag sets (embedding local extensions here provides
         a more convenient interface for those adding new tags:  if
         they define them in entity %TEI.extensions, they are embedded
         here and can therefore use our content-model parameter
         entities;  based on experience with Odd.dtd and P2X.dtd,
         I think this simplifies life a lot for the modifier.
 
TEIgis2.dtd
   contains a long list of entity declarations of the form
     <!ENTITY % n.blort 'blort' >
   if we substitute a different file, we can translate all gis into
   another language (but not attribute names, attribute values,
   or parameter entity names)
 
TEIclas2.dtd defines the major parameter entities used in content
models, including the entire class system of the core elements:
 
  I.  Low-level classes
      (as in published fascicle of CO, definitions of hqinter,
      hqphrase, etc., to seg)
 
  II.  High-level classes
      (more or less as in CO)
      <!ENTITY % m.phrase '...' -- all phrases, core or specialized -->
      <!ENTITY % m.inter  '...' -- all inter-level elements, whether
                                   in core or in tag set -->
      <!ENTITY % m.chunks '...' -- core chunks only; others go in
                                   tag-set-specific p.e.s -->
      <!ENTITY % m.common '...' -- core common and interlevel
                                   elements.  Would be defined as
                                   (%m.inter | %m.chunks) except
                                   that m.inter may have non-common
                                   members -->
 
  III.  Element classes marked for specific bases
        embed the %TEI.tagset.ent files, as appropriate, e.g.
 
        <![ %TEI.dictionary [
        <!ENTITY % TEI.dictionary.ent system 'teidict2.ent'>
        %TEI.dictionary.ent
        ]]>
 
        These files define the class ch.tagset and its model
        (e.g. %m.ch.dictionary), which contains all the chunk- and
        inter-level elements defined in the base (or:  unique to
        the base -- if gi sets are disjoint, this is the same thing).
        More below on contents of these.  Note that the general
        and mixed bases must embed all of these; it's not clear
        yet whether they will automatically embed all, or we can allow
        the user to specify exactly which bases should be included in
        a general file.  Either way, these files must not define
        chunk.seq -- instead they will define tagset.seq (e.g.
        prose.seq, dictionary.seq, like this:
 
        <!ENTITY % prose.seq '(%m.ch.prose | %m.common)' >
 
  IV.  Declarations of common content models
 
       <!ENTITY % phrase.seq '(#PCDATA | %m.phrase)' >
       <![ %TEI.prose [
       <!ENTITY % chunk.seq '%prose.seq' >
       ]]>
       <![ %TEI.dictionary [
       <!ENTITY % chunk.seq '%dictionary.seq' >
       ]]>
       <!-- etc. for all bases, but *not* for add'l sets -->
       <!ENTITY % paraContent '(#PCDATA | %m.inter | %m.phrase)' >
       <!ENTITY % specialPara '( (%m.chunk, %chunk.seq*)
                               | (%paraContent))' >
       <!-- note that this definition of specialPara requires
       a chunk.seq to begin with a *core* chunk, not a chunk
       unique to the base.  If this is a problem, user should
       redefine specialPara, or we should make it tagset-dependent. -->
 
TEIdict2.ent and other base.ent files
  declare any and all useful parameter entities, including at least:
     m.ch.dictionary (more generally m.ch.tagset, model group
             of the class ch.tagset).  This includes all the
             chunk and inter-level elements which are unique to this base
     dictionary.seq  (will become definition of chunk.seq -- this
             level of indirection is optional; a colleague persuaded
             me it would make the pattern of chunk.seq definitions
             in TEIclas2.dtd more obvious, and it also helps ensure
             that each fascicle and dtd file will define what it
             treats as a chunk.seq
     a.dictionary (attribute def of class dictionary) -- used in
             defining a.global (should not overlap with a.global; if
             it does, rules of inheritance will cause the a.global
             defs to win)
             if there are no global atts in a base, this is ''
 
TEItc2.ent and other additional.ent files
  similarly define useful parameter entities, but no m.ch.tagset
  or tagset.seq.  They *do* declare
 
  a.textcrit (attributes of class textcrit) - used in a.global
 
Tags defined by any tag set (base or additional) will all be in one
of the following groups, and will become part of the document grammar
as indicated.  I use the name 'tagset' as variable for 'prose' |
'verse' | 'drama' | ... | 'textcrit' | 'physical'
 
  phrases : phrase-level elements should be declared as members of
     one of the existing phrase-level classes (data, edit, hqphrase,
     loc, seg), OR declared as members of a new class, which is a
     member of class PHRASE.  The elements will then be picked up
     in the entity m.phrase, and will be legal in phrase.seq and
     paraContent.
 
  chunks : should be declared as members of m.ch.tagset; they will
     then be picked up as part of tagset.seq, and thus of chunk.seq
 
  inter-level elements : should be declared *both* as members of
     m.ch.tagset, which makes them part of tagset.seq and thus
     of chunk.seq; *and* as members of an existing inter-level class
     (bibl, hqinter, lists, notes) or of a new inter-level class,
     which will make them part of m.inter and thus of paraContent.
 
  others : elements which are none of the above may exist; they might be
     specialized subelements allowed only in certain crystals, etc.
     At some level, one of their legal ancestors must be chunk or
     inter-level element (or allowed as an inclusion exception on
     TEXT or TEI.2).