The Text Encoding Initiative's DTD has a liberal content model for the <TEXT> element (Sperberg-McQueen and Burnard 1994, 1192). Because the DTD requires <TEXT> at a high level, the relative liberality of the content model helps the markup scheme apply to a wide range of humanities documents. However, a data resource is only useful if the data are consistent in type of form and content. In the case of SGML-encoded texts, while a validating parser can check the form of the data the encoding project must enforce consistency in the type of content. The need for data consistency generates a principle of contextual equivalence, namely that an element <FOO> should always mark the same kind of content regardless of where that content occurs.° A related principle is that where two tags may be used for the same piece of content then a reason exists for using one rather than the other that has nothing to do with arbitrariness or convenience but comes instead from the nature or function of the piece of content. These two principles are fundamental to the logic of descriptive markup and they have particular relevance to the use of <TEXT>.Tagging a low-level object as a <TEXT> presupposes that object has a property or quality that makes it something more than just a paragraph, quote, line of verse, etc. According to the principle of contextual equivalence, the quality that makes the low-level object a <TEXT> is the same that makes the high-level object a <TEXT>. In effect, even though the low-level <TEXT> is optional while the high-level <TEXT> is obligatory, it is the possibility of a low-level <TEXT> that imposes a principled condition on what the high-level <TEXT> can be. The TEI element set is designed to reflect widely-shared textual concepts: broadly, everyone recognises paragraphs, quotes, dramatic speeches, etc. The association of <TEXT> with "text" is inescapable (whether the generic identifier is "TEXT" or "FOO"). While some uses of "a text" or "the text" do imply nothing more than an arbitrary piece of text (as in, for example, "the text for my sermon today is ....", or the text for a close-reading exercise), many other uses imply an identifiable thing: "a text" as a piece of text with a property or properties not shared by all other pieces of text. That the DTD specifically allows a low-level use of <TEXT> shows that TEI had this more specific concept in mind and wanted to accommodate it. The confusion engendered by this accommodation simply reflects the many different meanings assigned to "text" in everyday use. A workable description of what constitutes a text/<TEXT> must account both for the singularity of a text and for its ability to contain one or more things possessing a similar singularity. The TEI's Guidelines do not do this, so we must look beyond them. A promising place to start is the work of the Russian critic and textual theorist Mikhail Bakhtin, and with one essay in particular, "The Problem of Speech Genres." In this essay, written in the early nineteen-fifties but not published in Russian until 1979 and in English until 1986, Bakhtin distinguishes an utterance from a linguistic unit. Linguistic units--words, phrases, clauses, and sentences--are the building blocks of utterances, whilst utterances are the building blocks of communication. Utterances represent the smallest particles in a physics of language in use. Devoid of any communicative context, words, phrases, etc., are simply abstractions: representative examples of the formal properties of the language. Utterances, on the other hand, have unique properties deriving from their communicative context. Many times the boundaries of a linguistic unit and an utterance will coincide, but Bakhtin stresses that "[the] completely new qualities and peculiarities belong not to the sentence that has become the whole utterance, but precisely to the utterance itself" (Bakhtin 1986, 74). Utterances must consist of linguistic units, but linguistic units are not necessarily utterances. All utterances have a generic aspect, a form typical to the communicative context. Bakhtin calls these forms speech genres, and he distinguishes between primary and secondary speech genres. Secondary genres are more complex than primary genres because in "the process of their formation, they absorb and digest various primary (simple) genres that have taken form in unmediated speech communion" (Bakhtin 1986, 62). Simple kinds of utterance become compositional units of more complex utterances. For example, Bakhtin points to the genre of the novel which, because it represents human activity contains representations of various utterance types (letters, rejoinders in dialogue, etc.). The problem of "a text" versus "a piece of text," which is the problem of deciding what is a <TEXT> and what just a <DIV> or <QUOTE> or <LG>, lies in defining the qualitative difference between what may appear identical pieces of language. Bakhtin's distinction between utterances and linguistic units speaks directly to that problem, and his notions of utterances and primary and secondary speech genres offer an encoding project practical and principled criteria for using <TEXT>. Briefly, these are:

• <TEXT> applies only to the thing that is the utterance (eg. only to the epistolary novel itself, not to the letters that make it up)
• for a piece of writing to count as a <TEXT> there must be as much of it as is known to exist (with qualifications for different versions and historical contingency)
• when one instance of a secondary genre incorporates one or more other instances of secondary genres, and those incorporated instances would on their own qualify as <TEXT>s, then they should be tagged as <TEXT>s within the incorporating <TEXT>