T

The distinction Hoey mentions is made in this volume on practical, not theoretical grounds, and the overlap between text linguistics and discourse and conversation analysis should be borne in mind. Early modern linguistics, with its emphasis on

discovering and describing the minimal units of each of the linguistic levels of sound, form, syntax and semantics, made no provision for the study of long stretches of text as such; traditional grammatical analysis stops at sentence length. It is even possible to argue that ‘the extraction of tiny components diverts consideration away from the important unities which bind a text together’ (de Beaugrande and Dressler 1981: 21) and, although Zellig Harris (1952) had proposed to analyse whole discourses on distributional principles, employing the notion of transformations between stretches of text, this emergent interest in text and discourse study was lost at

notion of transformation to an intrasentential phenomenon. Early large-scale enquiries into text organi-

sation remained essentially descriptive and structurally based (Pike 1967; Koch 1971; Heger 1976), with occasional expansion of the framework to include text sequences or situations of occurrence (Coseriu 1955-6; Pike 1967; Harweg 1968; Koch 1971). Text was defined as a unit larger than the sentence, and the research was orientated towards discovering and classifying types of text structure; these were assumed to be something given, rather than something partly construed by the reader, and dependent on context. ‘We end up having classifications with various numbers of categories and degrees of elaboration, but no clear picture of how texts are utilised in social activity’ (de Beaugrande and Dressler 1981: 23). The descriptive method, however, tends to

break down because the language is too complex, with too many and diverse constituents to be captured. Ironically, it was the concept of transformations, lost by Harris to Chomsky, which allowed a new outlook on text that encouraged the upsurge in text linguistics during the 1970s. In transformational grammar, the infinite set of possible sentences of a language are seen as derivable from a small set of underlying deep patterns plus a set of rules for transforming these into the more elaborate actual surface structures. It was argued, first (Katz and Fodor 1963), that a whole text could be treated as a single sentence by seeing full stops as substitutes for conjunctions like and. This approach, however, deliberately leaves out reference to it

ignores the fact that ‘factors of accent, intonation, and word-order within a sentence depend on the organisation of other sentences in the vicinity’ (de Beaugrande and Dressler 1981: 24). This was noted by Heidolph (1966), who suggests ‘that a feature of “mentioned” vs. “not mentioned” could be inserted in the grammar to regulate these factors’. Isenberg (1968, 1971) lists other factors which could be dealt with within a single sentence, such as pronouns, articles and tense sequences, and ‘appeals to coherence relations like cause, purpose, specification, and temporal proximity’ (de Beaugrande and Dressler 1981: 24). Similar approaches to text analysis may be

found in the school of rhetorical structure analysis, where the emphasis is on how units of meaning (which are not necessarily sentences) relate to one another in a hierarchy, and how such devices as exemplification, summary, expansion, etc. build on core propositions to construct the finished text (Mann and Thompson 1988), an approach which in its turn owes much to the text linguistics of Longacre (1983). The Konstanz project, set up at the Uni-

versity of Konstanz in Germany, is related to these traditions of analysis. A group of researchers, including Hannes Rieser, Peter Hartmann, János Petöfl, Teun van Dijk, Jens Ihwe, Wolfram Köck and others, attempted to construct a grammar and lexicon which would generate a Brecht text; some of the results of this project are presented by van Dijk et al. (1972). The project highlighted more problems than it solved, though (de Beaugrande and Dressler 1981: 24): ‘Despite a huge apparatus of rules, there emerged no criteria for judging the text “grammatical” or “well-formed” … The problem of common reference was not solved’. The basic assumption of the undertaking was questioned by Kummer (1972), who points out that ‘the “generating” of the text is presupposed by the investigators rather than performed by the grammar’ (de Beaugrande and Dressler 1981: 25). In contrast to the grammatical method

employed by the Konstanz group, Petöfl’s (1971, 1974, 1978, 1980) text-structure/worldstructure theory (TeSWeST) operates with factors relating to text users rather than to the text as an isolated artefact, and with representa-

project is extremely complex (de Beaugrande and Dressler 1981: 25-6):

In the 1980 version, components are offered for representing a text from nearly every perspective. To meet the demands of the logical basis, a ‘canonic’ mode (a regularised, idealised correlate) is set up alongside the ‘natural language’ mode in which the text is in fact expressed. Rules and algorithms are provided for such operations as ‘formation’, ‘composition’, ‘construction’, ‘description’, ‘interpretation’, and ‘translation’. The reference of the text to objects or situations in the world is handled by a ‘world-semantics’ component; at least some correspondence is postulated between text-structure and world structure.