The WWP uses unnumbered <div> elements for high-level subdivisions of the text, and uses the type attribute to record the specific nature of the division. Current type attributes are limited to: act, advert, body, book, castlist, chapter, colophon, concluding, confession, contents, corrigenda, ded, endnotes, entry, epigraph, epilogue, essay, examination, frontispiece, index, letter, part, petition, poem, poemGroup, prayer, prefatory, prologue, recipe, revelation, scene, section, speech, subsection, undetermined, volume.
The use of the value “undetermined” should not be equated with the use of the <unknown> element. When the encoder encounters a piece of text that might or might not be a <div>, or whose type is uncertain but might be decided by someone else on a second pass, he or she should should tag it with the element <unknown> to mark it for further attention. However when the piece of text is definitely a <div> and its type is so obscure or so irrelevant that it does not warrant further thought, it should be tagged using this “undetermined” value.
Some of the type= values are derived from terminology used by the text itself to describe its structure (e.g. “revelation”, “examination”, “confession”). Where the text uses a special term for its major divisions the WWP may consider using a regularized, modern-spelling version of that term as the type= attribute value for those divisions, particularly (but not only) if that term seems likely to recur in other texts.
For the major generic subdivisions of prose, if the text itself does not suggest a terminology (if the divisions of the text are simply numbered, for instance, or separated by ornaments), we will use the standard modern terms for such divisions: part, chapter, and section, ordered as follows:
--one level of division: use chapter
--two levels of division: use chapter, section
--three levels: use part, chapter, section
--four levels: use part, chapter, section, subsection
--five or more levels: use additional “subsection” values nested recursively.
The rationale behind the decision to use special named terms where possible is that these terms are the ones which a user will rely on to retrieve textual components of interest, to limit the scope of searches, and possibly to establish comparisons between texts. Although it is unlikely that using these specific terms will enable comparisons across the textbase (since they aren’t likely to appear in more than one text), using a more general term would only allow a fairly meaningless comparison, so no functionality is lost by this resolution. Since these special terms arise fairly seldom, our previous concern that the DTD would be filled with idiosyncratic terms used in single texts has been set aside. It is always easy to convert more specific terms to more general ones if it turns out that we have zillions of weird values (reverie, soliloquy, fit, episode, bacchanal, etc.).