Annex 8

  Consistency

Excerpt from "Background and context for the development of a Corpus Encoding Standard"

Some inconsistencies are apparent within the TEI, which does not expound a precise set of principles for consistency. For example:

• In some cases, tag names refer to very generic categories and specific descriptions are put in attributes; in other cases, the tag name is specific. For example, for verse, hierarchical divisions are marked as <div1 type=canto>, <div2 type=stanza>, etc. (see Chisholm, 1995) In dictionaries, divisions are explicitly named with <entry>, <homograph>, <sense> (see Ide and Véronis, 1995b). The reason behind this is that for verse, an attempt was made to accomodate all relevant literary genres and cultures, whereas for dictionaries, all but western language dictionaries were explicitly excluded from consideration in order to enable a tighter definition of entry structure. In some other similar cases where consensus would exist in the large majority of cases, the tighter description is not accomodated: for example, although the division of book into chapters seems an obvious candidate, there is no <chapter> tag, and it is necessary to use <div type=chapter> (or <divn type=chapter>).

• For the most part, the TEI practice seems to be to include what might be regarded as a part of the text "content" as tag content, and to specify other material as attributes. However, the practice is not fully consistent: for example, in the encoding of spoken texts, certain semi-lexical items which are a part of the transcription content, such as "hmm" or "ha", can be encoded as attributes rather than tag content. Conversely, some material which is not a part of text content, for instance, notes and comments which might be made in the process of encoding the text, are included as tag content, even though they may not be considered to be a part of the text content proper.

• Most of the time tags identify structural or logical pieces of a text (e.g., paragraph, segment, utterance, date, etc.) of a text; sometimes they designate properties or processes (e.g., regularized, corrected, highlighted, emphasis, etc.). At other times properties are given in attributes. For example, one can say

  <name reg='Volange, de'>de Volanges</name>

  but not

  <w reg='though'>tho</w>

  (<w> marks a word, and "reg" indicates regularization). In the second case it is necessary to say

  <w><orig reg='though'>tho</orig></w>

  which is inconsistent with the first example.

  When tags are used to mark not only structural or logical objects in a text, but also to describe their properties, considerable verbosity can result. For example, if it is also necessary to indicate that the word is emphasized (<emph>), it is necessary to say

  <w><emph><orig reg='though'>tho</orig></emph></w>

  Note that in this example, there are six different permutations of the three tags <w>, <emph>, and <orig> allowed by the DTD to mark the same fact. Potentially all of them could appear in the same document using the TEI DTDs. This seriously weakens the structural model of the text, since at a certain level virtually anything can contain anything, thus creating problems for validation, search, etc.