Today, ‘text’ is widely defined as an empirical communicative event given through human communication rather than specified by a formal theory. Each such event ‘rides on’ a dynamic dialectic between the ‘virtual system’ of language (the repertory of possibilities) and the ‘actual system’ constituted by the choices of the text producer; the text is thus on neither side of language versus use, but integrates and reconciles the two. This paper sets out to describe, as empirically and realistically as it can, the processes whereby communicative participants can and do produce and receive texts. The task plainly demands interdisciplinary research between text linguistics and psychology, sociology, ethnography, and so on, all of whom work with real data from the standpoint of human activities. Defining ‘text’ and the paper's main task this way brings ‘text linguistics’ into line with the research tradition of discourse analysis, which has operated ‘bottom up’ with realistic data obtained from fieldwork. Since its earliest formulations (e.g. Harris 1952), investigators have looked a real data beyond the sentence and developed a ‘grammar’ incorporating semantic and pragmatic issues, such as how different languages deploy grammatical resources for expressing events and participants in narratives (see now Longacre 1990). The ethnographic emphasis obviated the need for the abstract theoretical formalisations cultivated in ‘mainstream linguistics’, including the isolation of ‘language in and for itself’ from the use of language (cf. Stubbs 1993; Sinclair 1994). Today, the same emphasis has brought discourse analysis into the mainstream, while formalist linguistics drifted into a ‘scientific crisis’ by failing to meet its announced goals, notably to formulate a complete ‘generative grammar’ for any natural language. The constraints once excluded on theoretical grounds now had to be reintroduced, and discourse analysis steadily merged with text linguistics. The latest and most decisive step has been the compilation of the ‘Bank of English’, a vast computerised corpus of contemporary spoken and written English discourse, which enables reliable statements about the typical patterns and collections of English (Sinclair 1992a, 1992b; Baker et al. eds. 1993). The implications Sinclair (1994) draws from such data are likely to be felt through all of linguistics. Text linguistics too will feel the impact. First, the dialectic between virtual system and actual system becomes far clearer and more direct when the function of a word of phrase in the representative corpus can be explicitly correlated with the function it has in a given text. Second, the text as whole does not need a formal circumscribed definition; for most purposes, it can be perfectly well grasped as one actualization within a set of attested cross-sections we can pull up for display as we choose. Third, the sentence boundary loses its theoretical significance and becomes an empirical phenomenon which, like the other boundary signals of punctuation or intonation, occurs with higher probability after some word combination (Firth’s ‘collocations’) and grammatical patterns (Firth’s ‘colligations’) than others. Fourth, the production of a text is viewed as a set of interactive local choices, some of which are made well before the key word appears in the sequence. And fifth, subcorpora can be used as a data-driven bottom-up method for characterising text types and discourse domains, a task for which speculative schemes have remained ineffectual. It is now possible to reach a consensus on realistic definitions of the two main terms: text being the individual communicative event within the ongoing interaction of discourse. Shared research interests should decide how abstract or concrete, how theoretical or empirical, these terms should be in specific projects with explicit goals. Text research is also swinging into line with discourse processing, an interdisciplinary exploration of the processes that make discourse both feasible useful in so many settings. Further, text linguistics has become strongly engaged with pragmatics (e.g. Schmidt 1973, 1976). Instead of giving a set of ‘rules’ that “turn these objects into acts” and “place these acts in a situation” (van Dijk 1976:), pragmatics investigates how texts and co-texts (not sentences) are always already acts in situations, even while the syntactic or semantic operations are being carried out. Hence, pragmatics forms the outermost framework for approaching syntax and semantics in terms of communicative activities. And finally, text linguistics has assumed a cordial relation to stylistics (e.g. Enkvist 1977), which mainstream linguistics tended to marginalise by situating it in the use of language rather than in language itself. The increasing engagement of text linguistics with the four domains just cited — discourse analysis, discourse processing, pragmatics, and stylistics — has both increased this discussion's flexibility and guided it in resetting its priorities. In retrospect, then, one might distinguish three stages in the rise of ‘text linguistics’. The earliest stage (late 1960s, early 1970s) was dominated by ‘text grammar’ and remained fairly circumscribed and uniform by working to extend prevailing theories from sentence to text. The next stage (late 1970s, early 1980s) was more dominated by ‘textuality as ‘a structure’ with ‘both linguistic and social aspects’. The ‘principles of textuality’ apply whenever an artifact is treated as a text; so textuality is not a linguistic property or feature (or set of these) which some samples possess and other do not, but an empirical human predisposition and activity wherever communicative events occur. The mutual relevance assigned to linguistic forms is cohesion and that assigned to the ‘meanings’ is coherence; intentionality covers what speakers intend, and acceptability what hearers engage to do; situationality concerns ongoing circumstances; and intertextuality covers relations with other text, notably those of the same or a similar ‘text type’. The third and most recent stage has been dominated by ‘textualisation’, the social and cognitive processes entailed in the actual production and reception of texts. This trend too was driven by the search for richer and more global constraints, but now assuming that ‘process models’ can provide more than can ‘product models’ starting from a text-artifact. The 1990s look toward a general science of text and discourse, merging its ‘syntax’, 'semantics', and 'pragmatics' in a framework for exploring the relation of texts to such factors as text types, style, register, language for special purposes (LSP) and terminology, but also socialisation, education, translation, ideology, gender, and emotion.