Text linguistics at the millennium:

Corpus data and missing links 

robert de beaugrande  

Abstract  

Text linguistics seems to have originated chiefly in order to expand the search for constraints, which was being noticeably impeded by the self-imposed restrictions in a ‘linguistics’ centred on isolated, invented sentences and abstract formalisms. Yet early attempts to bring the ‘text’ into the scope of su ch a linguistics now seem inverted: for us the actual text, not the invented sentence, must be the essential linguistic unit, and is sustained by internal systemic organisation and by its external systemic organisation within one or more ‘intertexts’. In the coming millennium, this prospect can now finally be documented and clarified by working with very large corpora of authentic texts, whereby we can hope to uncover some of the vital and delicate missing links between ‘language’ and ‘text’. 

A. ‘Language’ and ‘text’ in ‘modern linguistics’ 

1. Modern ‘general linguistics’ has been a singular enterprise, at times most sharply distinguished from other approaches to ‘language’ by its self-imposed restrictions. Since its outset, it has been influenced by Saussure’s (1966 [1916]: 232) aspiration that ‘the true and unique object of linguistics’ should be ‘language studied in and for itself’. In effect, this vision of ‘linguistic’ science has been pursuing the question: what would ‘language’ look like when it’s off by itself and not being used (cf. § 14, 82)?

2. An unwelcome answer would be that it no longer looks like a real language. If so, the term ‘language’ used by this linguistics loses its ordinary meaning, namely: a mode of communication used among the members of a human community. Some common uses in this meaning can be seen in these authentic data supplied in July 1994 from the ‘Bank of English’ at Birmingham University, the world’s largest computerised text-data corpus, then containing over 200 million words (cf. § 60, 64, 89):

(1)   you should be pleased that the French language has been spared

(2)   he has no qualifications in teaching English as a Foreign Language

(3)   it’s old-fashioned, and it’s in a foreign language. People are frightened of it

(4)   I was told afterward that my language was most entertaining        

(5)   fatally damaged? I don’t want to use language of that sort

(6)   there is a lot of bad language and gratuitously oafish behaviour

(7)   General Kryuchkov used the language of the Cold War when he accused the US

(8)   He violently opposes the new language law, which makes major concessions to ethnic minorities

Sample (1) concerns the ‘French language’ spoken as a native language by a whole nation, whereas sample (2) concerns the ‘English language’ as a subject-matter to be ‘taught’ to, and learned by, people who speak a different native language. Sample (3) implies that ‘a foreign language’ may contribute to ‘frightening people’. Samples (4), (5), and (6) indicate how particular uses of ‘language’ get evaluated. In sample (7), ‘language’ covers both the style and the content of what a Russian General said — belligerent in tune with the ‘Cold War’. And sample (8) mentions a government regulation concerning which language or languages should be used and when, for instance as an ‘official language’ in a multilingual country. So all these uses relate to real people who either might use a ‘language’ for communication, or else might be hindered in doing so because they had ‘no qualifications’ (2), or because the ‘language’ was ‘foreign’ (3) or was restricted by a ‘language law’ (8), and so on. None of the uses matches ‘language in and for itself’ in the austere theoretical meaning of Saussurian linguistics.

3. This mismatch might explain why Saussure (1966 [1916]: 9, 11) asserted that ‘speech cannot be studied’, ‘for we cannot discover its unity’; it is only a ‘heterogeneous mass’ of ‘accessory and accidental facts’ (§ 10, 21f, 39, 82). Later, Chomsky (1965: 4, 201) asserted in similar vein that the ‘observed use of language’ ‘surely cannot constitute the subject-matter of linguistics, if this is to be a serious discipline’; ‘much of the actual speech observed consists of fragments and deviant expressions of a variety of sorts’. As if in parallel, these same linguists asserted that ‘the concrete entities of language are not directly accessible’ (Saussure, 1966 [1916]: 110); and that ‘knowledge of the language, like most facts of interest and importance, is neither presented for direct observation nor extractable from data by inductive procedures of any known sort’ (Chomsky, 1965: 18).

4. The mismatch is thus a signal that ‘language’, being the true ‘subject-matter of linguistics’ as a ‘serious discipline’, is not what real people hear or see in ‘actual speech’. Another well-known claim then falls into place: ‘linguistic theory is concerned primarily with an ideal speaker-listener, in a completely homogeneous speech-community, who knows its language perfectly’ (Chomsky, 1965: 3). Neither this ‘speaker’ nor this ‘community’ ‘exist in the real world’, as Chomsky (1977: 172) has calmly conceded. So the true nature of ‘language’ is known only to academics whose degrees in ‘theoretical linguistics’ somehow equip them with privileged access to ‘perfect knowledge’ (cf. § 29, 34, 83).

5. One of Saussure’s (1966: 8) most candid acknowledgements also falls into place: whereas ‘other sciences work with objects that are given in advance’, in ‘linguistics’ ‘it is the viewpoint that creates the object’. What he did not acknowledge but has been extensively displayed in the subsequent history of modern linguistics is that multiple ‘viewpoints’ create multiple ‘objects’, whereby the meaning of the term ‘language’ has grown steadily more unstable and obscure. Linguistics has been fragmented into disputatious factions, each assigning to ‘language’ its own idealised meaning. And much time and print has been expended upon disputing over whose idealisation is better without turning for adjudication to the evidence of real language in ‘actual speech’ (Beaugrande, 1997a, 1998a, 1998b).

6. What then about the status and meaning of the term text? In the field proposed by linguists from Saussure up through Chomsky, the ‘text’ would seem to merit no home at all. As Halliday (1994: xxii) has remarked, Saussure’s ‘understanding of the relationship between the system of language and its instantiation in acts of speaking’ implied that ‘the text’ ‘can be dispensed with’; and ‘linguistics, for much of the twentieth century’, has accordingly been ‘obsessed with the system at the expense of text’

7. A few linguists did proffer a home for the ‘text’ in their theories, but failed to secure it in their own practices of investigation. For Hjelmslev (1969 [1943]: 12), ‘the linguistic investigator is given’ ‘the as yet unanalysed text in its undivided and absolute integrity’ (but cf. § 88). So ‘linguistic theory starts from the text as its datum and attempts to show the way to a self-consistent and exhaustive description of it through an analysis’ (1969: 21). The ‘theory’ must also ‘indicate how any other text of the same premised nature can be understood in the same way’ by ‘furnishing us with tools that can be used on any text’ (1969: 16). ‘Obviously, it would be humanly impossible to work through all existing texts’, and ‘futile’ as well, if ‘the theory must also cover texts as yet unrealised’ (1969: 17) (cf. § 88). Even so, Hjelmslev counselled that ‘linguistic theory’ should seek to ‘describe and predict’ ‘any conceivable or theoretically possible texts’ ‘in any language whatsoever’ (1969: 16f). He even stipulated that ‘linguistic theory’ should ‘foresee’ ‘a language without a text constructed in that language’ (1969: 39f) (cf. § 37).

8. Hjelmslev’s line of argument implied a programmatic move to situate the ‘text’ inside the same abstract theoretical sphere as ‘language’ (or ‘langue’) defined by Saussure, whose programme Hjelmslev is known to have heartily approved. We might sense a remarkable expansiveness in the aspiration for ‘linguistic theory’ to provide one single method of ‘description’ and ‘analysis’: first for the one ‘unanalysed given text’, then for ‘all theoretically possible texts’, and finally even for the non-existent texts of ‘a language without a text’. How this aspiration could be put into practice is obscure: in the four published volumes of Hjelmslev’s writings, I could not find even one demonstration. His only directive was that ‘the text is regarded as a class analysed into components, then these components as classes analysed into components, and so on until the analysis is exhausted’ (1969: 12f). So the text would be like a big chunk of language waiting to be taken to pieces which would then be taken into smaller pieces, over and over, until we arrives at the opposite end from its ‘undivided integrity’ (§ 7, 57).

9. We might contrast Hjelmslev’s programme with that of J.R. Firth, who declared that ‘the text is the main concern of the linguist’ (1968 [1952-59]: 24, 90, 173). ‘All texts’ ‘in modern spoken languages’ are considered to ‘carry the implication of utterance’ and are ‘referred to typical participants in some generalised context of situation’ (1957 [1951]: 220, 226). So the ‘attested language text’ should be ‘duly recorded’ and ‘abstracted from the matrix of experience’ (1968: 199f, 99). Since it may not be ‘possible or desirable to present the whole of the materials collected during the observation period’, some ‘corpus’ is ‘essential’ (1968: 32) (cf. sections E, F, and G).

10. Still, Firth did not intend to countermand Hjelmslev’s programme by situating the text on the side of ‘speech’ in opposition to Saussure’s ‘language’, because Firth (1968: 28, 41, 127, 139f) roundly rejected the whole dichotomy between ‘langue and parole’. By ‘referring’ the text to ‘typical participants’ and ‘generalised contexts’ and by ‘abstracting it from the matrix of experience’, the linguist would lift the text up from the merely ‘heterogeneous, accessory, and accidental’ plane where we saw Saussure situating ‘speech’ (§ 4). Unfortunately, Firth’s own four published volumes give only a few sketchy demonstrations, far short in practice of what his writings on theory projected.

11. At all events, the ‘text’ was likely to remain on the margins of linguistics as long as it was eclipsed by the sentence. Curiously, the ‘sentence’ did not start out as the prototype of ‘language’ in the meaning of Saussure. He had argued that although ‘the sentence is the ideal type of syntagm’, ‘it belongs to speaking [parole], not to language [langue]’ (1966 [1916]: 124). However, his reservations distinctly placed the sentence on both sides: ‘in the syntagm there is no clear-cut boundary between the language fact, which is a sign of collective usage, and the fact that belongs to speaking and depends on individual freedom; in a great number of instances it is hard to classify a combination of units because both forces have combined in producing it, and they have combined in indeterminate proportions’ (1966: 124f; see Beaugrande, 1999a for discussion).

12. These early reservations did not prevent the sentence from later occupying the centre of modern linguistics, most famously when the ‘generative’ approach defined a ‘language’ to be an ‘infinite of set sentences’ (Chomsky, 1957: 13) (cf. § 32). Officially, the ‘sentence’ was a unit purely restricted to ‘syntax’ or ‘grammar’, two terms now used as if they were interchangeable, although they cannot be, as we shall see (§ 17f). Yet this restriction could not be sustained during attempts to actually formulate a ‘generative grammar’, even for small samplings of invented English sentences. The term ‘sentence’ was also being unofficially used (among other things) for a semantic unit that should be called a ‘proposition’ and for a pragmatic unit that should be called a ‘speech act’, as if to paper over the ‘indeterminate proportions’ in Saussure’s reservation (Beaugrande, 1997a). So just when ‘generative’ linguists were volubly announcing that the ‘study of competence abstracts away from the whole question of performance’ and ‘why speakers say what they say, how language is used in various social groups, how it is used in communication, etc.’ (Dresher and Hornstein, 1976: 328), the concept of the ‘sentence’ was being quietly stretched and bent in the direction of ‘performance’ (cf. § 50).

13. As long as the ‘sentence’ was getting stretched, the need to recognise the ‘text’ might not have seemed urgent. But the restrictions of the single sentence eventually had to become onerous for linguistic investigations even in the narrowed area of ‘syntax’ or ‘grammar’. And some linguists would eventually respond by looking ‘beyond the sentence’ and at the ‘text’. Not surprisingly, much early work in ‘text linguistics’ emblematically aspired to build upon what now came to be called, for purposes of contrast, ‘sentence linguistics’ — a term that would have seemed oddly redundant to the ‘generative’ approach. By exploiting the ‘indeterminate proportions’ implicit in the sentence (§ 11), such work need not settle the question of whether the ‘text’ might be a unit of the ‘actual speech’ rejected by the linguists who had followed Saussure and Chomsky (cf. § 3, 10, 23).

14. The most straightforward strategy for getting the text inside a linguistics designed for the sentence would be to define the ‘text’ as a sequence of sentences. As such, the ‘text’ could smoothly inherit the established properties of the sentence, such as being ‘grammatical’, ‘well-formed’, and ‘rule-governed’. Yet disputes arose in the early 1970s over the question of whether those properties were sufficient, so that established ‘sentence linguistics’ could account for whatever might be found (Dascal and Margalit, 1974); or whether new properties would be found that belong or apply only to texts (van Dijk, 1972).

15. Text linguists naturally favoured the second option, which justified their work as a accredited field. But in retrospect, I find the question inverted. Following such leads as Hartmann (1968, 1971) and Schmidt (1973), we can accept the text as the essential linguistic unit and then explore the status of the sentence as one of the potential segments within a text, and thus one that could explicitly benefit from the previously implicit bending in the direction of performance (§ 12, Beaugrande, 1999a). Moreover, the sentence could richly inherit the properties that have since been established for the text, such as being ‘cohesive’, ‘coherent’, ‘intentional’, ‘acceptable’, ‘informative’, and ‘situational’ (Beaugrande and Dressler, 1981) (cf. § 59).

16. In the process, we might reach the unsettling conclusion that syntax doesn’t exist in natural language insofar as the term means a system of ‘formal rules’ for arranging words together in sentences (cf. § 14) (compare García, 1979; Givón. 1979). As already implied by Saussure’s reservation about the ‘sentence’ (§ 11), speakers and writers certainly select and combine words in response to important ‘non-syntactic’ factors, such as lexical preferences, communicative situations, and personal motivations (§ 98-101). To assert that these factors are all ‘beyond the sentence’ or ‘outside the sentence’ is to ignore their potent effects inside the sentence and to block off significant resources for description and analysis.

17. The restrictions of syntax become most obtrusive when it gets directly equated with ‘grammar’ (§ 12). Despite the once fashionable notion of a ‘deep structure underlying surface structure’ (e.g. Chomsky, 1965), a ‘grammar’ which looks only at the order of words inside isolated sentences and not at ‘why speakers say what they say, how language is used in various social groups, how it is used in communication’ (§ 12) must remain a shallow and superficial enterprise. In English — unlike many other languages — word-order is sufficiently ‘frozen’ in some areas to sustain carefully selected Aspects of the Theory of Syntax’; and these have indeed been the concerns of ‘generative linguistics’ until it withdrew into ‘universals’ and ‘mental representations’, where the ‘sentence’ is no longer crucial (see Beaugrande, 1998a for discussion).

18. A realistic and empirically justified ‘grammar’ would rather be the front end for all the relevant motivations that speakers or writers typically apply or reflect when they put classes of words (e.g. Nouns) or of word-parts (e.g. Suffixes) in one order rather than another (cf. § 101). This lesson might also be drawn from the recalcitrant obstacles encountered by the early projects of text linguistics to construct a text grammar from the top down in the narrow and abstract ‘generative’ sense. The ‘text grammar’ for a fairly short text by Bertolt Brecht (‘Herrn K.s Lieblingstier’), which even shows a simple vocabulary and some parallelisms in its phrasing, got tangled in an explosive complexity of ‘rules’¹ (cf. van Dijk, Ihwe, Rieser, and Petöfi, 1972), and the project was eventually abandoned with no official conclusion (cf. § 38). The same fate can be predicted for any project to describe ‘texts’ on comparable levels of abstraction and formality: what gets ‘abstracted away’ during the ‘formalisation’ would be vitally needed in order to account just for the order of words, and far more for the choices of words (cf. § 40).

19. The converse approach for a ‘text linguistics’, and the one that has gradually won out, is to work from the bottom up. We need to examine a comprehensive range and variety of authentic texts and explore what sorts of properties deserve to be accounted for, including, but not restricted to, those of ‘grammar’ in the broad sense of § 18. We can apply whichever categories and concepts of previous ‘linguistics’ seem productive, but we can also apply ones from adjacent fields, such as literary studies, cognitive science, artificial intelligence, ethnography, economics, and political science (cf. Beaugrande, 1980, 1997a, in preparation) — whatever bases we can enlist in exploring how speakers and do select and combine words inside phrases, clauses, sentences, or any other relevant units, such as paragraphs, essays, or science textbooks.

20. This approach has a distinctly Firthian flavour, although we can exploit fields, methods, and resources that were not available to Firth and his pupils. Today, we are far better positioned to ‘observe’ and ‘collect’ a ‘corpus’ of ‘attested language texts’ and to determine what is ‘typical’ and ‘generalisable’ (§ 9f). Admittedly, the sheer size of the task remains daunting enough to indicate why Saussurian linguistics was so eager to rule out the ‘study of speech’ (§ 3). Yet the task deserves to be at the top of our agenda for the next milennium. 

B. Virtual system and actual system  

21. As a provisional strategy, some ‘text linguists’ (myself included) have for a number of years been proposing to view the relation between a language and a text as one between a virtual system and an actual system. One pioneer for this view was the eminent linguist and language philosopher Peter Hartmann,² who, together with his pupils (e.g. Siegfried J. Schmidt, Roland Harweg, Walter A. Koch, Götz Wienold), were the most thoughtful originators of text linguistics. Against any linguistics that might view texts being like Saussure’s ‘speech’, i.e., ‘heterogeneous’, ‘accidental’, and devoid of ‘unity’ (§ 3), we maintain that the text is internally systemic on its own terms and is externally systemic in respect to other texts (§ 47, 99). The text is thus an intersystemic event during which multiple systems interact and converge (§ 89).

22. This viewpoint does not mean that ‘the interplay of langue and parole somehow vanishes on the level of texts’ — a consequence inferred from my work by Lindemann (1981: 126). Much in the spirit of Firth (§ 10), text linguistics firmly rejects the dichotomy of ‘langue and parole’ for having been deeply misconceived all along. The major flaw, not widely recognised, has lain in attributing to ‘language’ an ideal order, and to ‘speech’ an accidental disorder (Beaugrande, 1998a, 1998b, 1999a). The bizarre implication would be that using a ‘language’ in ‘speech’ triggers an abrupt catastrophic transition from stable and integrative order over to unstable and disintegrative disorder. Conversely, the linguist whose ‘study abstracts away from how language is used in communication’ (§ 12) miraculously restores the pristine, ideal order. The practices of linguistic investigation would consist chiefly of ‘idealising language’; yet we should be wary of projecting an ideal that is fully disconnected from real language (Beaugrande, 1998a, 1998b) (§ 97).

23. Nor again does this same viewpoint mean that ‘texts seem to be exclusively created by an actualisation of lower-level virtualities’, and that ‘texts do not dispose of any virtual systemic aspects of their own’ — further consequences inferred by Lindemann (1981: 126). Of course texts hinge crucially upon the ‘virtual systemic aspects’; our problem is that ‘virtuality’ is not readily isolated from ‘actuality’ when we work with real texts. Lindemann himself proposed that ‘actually occurring texts may be virtualised by exploring’ their ‘systemic aspects’, yet this is precisely what text linguists do in practice though without describing our practices by that term. However, we cannot be just putting ‘virtuality’ back into a space where it has been ‘actualised out’, because texts always retain some aspects of their virtuality as long as the language is still known and used.³ So the ‘text’ is not just a unit of ‘actual speech’ (‘parole, langage, performance’ etc.) in the senses of Saussure and Chomsky (cf. § 3, 10, 13), but is rather a unit which actually links language to speech.

24. Some powerful evidence for the ‘virtuality’ would be that a single text can be received and interpreted (‘actualised’) in more than one way by different participants in a text-event or even by the same participant at different times. Such is indeed a constitutive principle of literature; the literary text might, as Wellek and Warren (1956: 152) have done, be compared to ‘langue’, and each ‘individual realisation’ to ‘parole’. Yet far from navigating between the order of ‘langue and the disorder of ‘parole’ (§ 22), the audience recreates a novel order for the ‘textual world’. Your motivation a reward for reading or listening to a literary text is thus to be a privileged participant each time in constructing its ‘world’, whose order is in part your own creative achievement (Beaugrande, 1988). The same process would apply, on a far less conscious plane of creative awareness, for the actualisation and re-actualisation of any text; literature merely accentuates and thematises the process to deepen and broaden our understanding of the human situation.

25. The implication would be that the ‘same’ text need not have the ‘same order’ for everybody. At least in its fine details, each actualisation is unique and unrepeatable. Moreover, each participant has a partially individualised knowledge of the virtual system, with greater variation in the lexicon and lesser variation in the grammar (§ 83ff). So we must resolutely address the questions of how and how far the respective actualisations of any text can agree or coincide well enough for a text receiver to feel confident of ‘understanding’ what the producer ‘meant’; or for several participants to receive a text in what is thought to be the ‘same way’ Evidently, the degree of convergence among participants in a text-event is achieved on line through the interaction of multiple systems whose design both anticipates and adjusts to these achievements.

26. The production and reception of a text would be complementary transitions between a more open mode and a more closed mode of systemic order; and these two modes would determine each other in a continual dialectic. Contrary to the assumptions of linguistics cited above, we would conclude that these two modes of order must be quite proximate: the order of language is more closed and the order of texts more open than has been widely acknowledged before (§ 56).

27. Text linguistics accordingly needs to develop models of how the one mode of order sustains and tunes the other. We might postulate a cycle whereby the language ‘actualises’ to sustain and tune the texts, whilst the texts ‘virtualise’ to sustain and tune the language. My intransitive usages of the two verbs may sound esoteric, since the agents of these reciprocal activities are of course the participants in the text-event; but the activities are normally just by-products of communicative interaction and do not match what the participants are consciously doing or intending to do (Hartmann, 1963). The exceptions would of course be the poets, whose enterprise consists of seeking and testing new ways of organising language (Beaugrande, 1978, 1988).

28. Our own enterprise as text linguists consists of consciously ‘virtualising’ texts whenever we ‘actualise’ them for the purpose of drawing inferences about the order of the virtual language system (§ 23). A key question, still to be adequately explored, is our own version of the well-known problem of ‘participants’ versus ‘observers’ (cf. Harweg, 1968). How can far our conscious and attentional ‘actualising-for-virtualising’ correspond to, or accurately describe, the unconscious and automatic actualising and virtualising of ordinary text producers and receivers; and what might occur during our processes of making conscious and focusing attention? Quite plausibly, the potential of the texts being investigated gets considerably more elaborated precisely insofar as we prolong and intensify the cycle postulated in § 27 (see also § 83ff).

29. In addition, the principle of each participant having a partially individualised knowledge of the virtual system (§ 25) must apply to us too. What might be our status as individuals and within the wider community? Even if we could set aside our academic training and consider ourselves reasonably typical, we could not yet consider ourselves representative, due to several constraints. The most obvious constraint is that the language experience of any individual, even one who has read or written extensively, is only a small part of the experience of the community. Instead of conveniently taking the community to be ‘homogeneous’ and purporting to have privileged access to the ‘perfect knowledge of the language’ (§ 4), we need to lay open the degrees of real uniformity and diversity among language users and among their texts to large-scale adjudication with substantive evidence.

30. A less obvious constraint on being representative could be inferred from the virtuality I have suggested that texts always retain (§ 23). This cannot be directly measured by their manifested uniformity, because ‘virtuality’ is by definition open and dynamic, and what is manifested is the actuality that partly confirms it and partly specifies or modifies it to suit the context. Instead of vowing that ‘knowledge of the language’ is never presented for direct observation nor extractable from data’ (§ 3), we can assume that ‘knowledge of the (virtual) language’ never totally converges with any set of actualities, but does move steadily closer toward convergence as the set gets larger and more diversified. In that sense, we all — text linguists or not — remain ‘language learners’ throughout our lives as we accumulate experience with texts. Text linguists are just exceptionally self-conscious ‘learners’, ‘observing’ and ‘extracting’ whatever we can in the knowledge that there is always far more beyond it all.

31. Still another constraint on being representative is the unresolved uncertainty about intuition providing the source for a ‘grammar’ to ‘describe the intrinsic competence of the idealised native speaker’ (Chomsky, 1965: 24). Contrary to such well-known claims, unaided intuition is not broader and deeper than language experience but narrower and shallower. It is best secured for the more stable and general ‘frozen’ areas of the virtual system whose virtuality seems static and hence independent of ‘actual speech’, whence the preference for restricting linguistics to a few Aspects of the Theory of Syntax (cf. § 17). But intuition is not well secured for the dynamic cycles of actualising and virtualising precisely because of the continual tuning. The interacting systems settle down in precise detail only at the respective stages of the interaction itself. So the native speakers’ intuition may well not be reliable for telling you what they would say until there arises a real-life occasion when they do say it. Intuition operates most smoothly after the fact in making sense of what has already been said (§ 99).

32. This overall line of argument also indicates that the language system never settles down in a ‘synchronic’ dimension; consequently, all attempts since Saussure to construct a ‘synchronic description’ are doomed to be partial, restricted, and provisional. However, these limits should not disband the ‘synchronic’ linguists so much as rather resituate their enterprise. Like total convergence (§ 30), ‘synchronicity’ is a factor we can move closer to yet never attain. It is not set apart in some ideal ‘language’ disconnected from ‘speech’ but represents the totality of simultaneous actualisings of the virtual system (cf. Harweg, 1968: 142). So we cannot remotely achieve any ‘synchronic description’, however partial, by ‘dispensing with texts’ (cf. § 6); we need bigger and better means for examining what large sets of texts have in common.

33. We thus return to a deep question: if language and text, or virtual and actual, are closely interconnected yet never converge, how then are their connections sustained? On a highly general plane, we might stipulate: the language is a set of texts; each text is one member of the set. We might interpret this stipulation by viewing a ‘language’ as one vast ‘continuous text’, as ‘the totality of what has been written and spoken (or perhaps even thought) in a language’ (Harweg, 1968: 142, my translation).

34. We might feel reminded here of the definition (already cited in § 12) of a language being ‘an infinite of set sentences’ (Chomsky, 1957: 13).⁴ The term ‘infinite’ merits closer examination, as I have shown elsewhere (Beaugrande, 1998a). The ‘infinity’ was hypothesised on the simplistic mechanical basis of ‘recursive devices’ (Chomsky, 1957: 23f) whereby any sentence could, in theory, be made longer or more complex. In practice, the set of infinitely long or complex sentences is an empty set and thus an intractable object for constructing a ‘grammar’. Nor can it plausibly be attributed even to the ‘perfect knowledge’ of ‘ideal speakers’, who also form an empty set of humans (or superhumans) (cf. § 4).

35. Still, for the sake of argument, we could infer that if a text is defined to be a ‘sequence of sentences’ (§ 14), then ‘generative’ theory would allow for an infinitely long text. That text would be theoretically equivalent to one superlong sentence with ‘sentence boundaries as sentential connectives’ (Fodor and Katz, 1964: 491). Yet such a text would also inescapably belong to an empty set. At most, our theorising could plausibly postulate the concept of an infinitely long intertext, whose practical correlate could be a conversation in which the ‘last word’ can never be said. Such a notion has a ‘post-modern’ or ‘post-structuralist’ flavour, though we are arguably still dealing with an empty set. What is given is a real set of intertexts such as conversations, each of which is at any one moment finite, though quite possibly still open, such as the discourse of history, which could be definitively terminated only if history came to an end.

36. Halliday (1997: 6) has for some time ‘preferred to reverse the principle and characterise a language as an infinite system generating a finite body of text’; but following a recent discussion where I pointed out that, in terms of mathematical theory, the ‘infinite’ must include combinations with infinitesimal probabilities, he now favours ‘replacing “infinite” with “indefinitely large”’. The finiteness of the set of texts in existence at any given moment in time is hardly disputable, but is a weak constraint for a living language, especially one spoken as widely as English. Another weak constraint comes from the ‘indefinite largeness’ of the set of potential texts in English. The key question, as pointed out elsewhere by Halliday (1993), remains: how does the set of existing texts constrain the set of potential texts such that, at any given moment, some ways of adding to the existing set are substantially more probable than others?

37. Problems with sets also arose when, influenced by the high-level aspirations of ‘generative linguistics’, some text linguistics essayed to formulate the constraints for distinguishing the set of texts from the set of non-texts (see Dressler, 1972 for discussion). At the time, these constraints seemed intuitively plausible in theory, but we can now see why they proved intractable in practice. The attempt to create a non-text is always just one step within an ongoing textual event, and thus still a sub-text, however unconventional. Its ‘textuality’ is locally disrupted but globally sustained. So in both theory and practice, the ‘set of non-texts’ is one more empty empty set and therefore obeys no constraints. Nor can it be identified with the set of non-existent texts for which Hjelmslev would have us postulate a ‘language’ (§ 7f), since those would still be ‘texts’, although virtual — texts of a ‘language’ whose ‘textual process’, i.e., its actualisation, is itself purely ‘virtual’ (cf. Hjelmslev, 1969 [1943]: 40).

38. I shall cite just one more constraint on our being representative: the practices of text linguistics who analyse and describe texts inevitably add to the set of texts. We thus incur the additional challenge of probing the relations between the set of texts we are analysing and the set of texts we are producing as we go along (cf. § 83ff). This challenge corresponds to Firth’s (1968) distinction between ‘language under description’ and ‘language of description’, and to his epigram about linguistics being ‘language turned back on itself’. Evidently because science in general has preferred to regard its own texts ‘as a purer code, eschewing rhetoric and simply reporting natural fact’ (Bazerman, 1988: 6, 14), this challenge is still far from fulfilled. My own studies have consistently found that the relation between scientific texts and their domain they purport to account for is complex and unstable, especially when that domain is language itself, as in modern linguistics. Far too little attention has been devoted to the practical strategies of a linguist producing and receiving a text about ‘language’, even though they have significant effects upon the operations and results of ‘theorising’ (Beaugrande, 1984, 1991).

39. For the present discussion, one effect I would highlight has been the widespread textual strategy among linguists of inventing sets of artificial pseudo-texts, usually isolated sentences which are either trivial in their ‘grammaticalness’ (like ‘the man hit the ball’) or else wildly bizarre in their ‘ungrammaticalness’ (like ‘*ball the hit ball man the’). These are intended to capture some constraints, optimistically called ‘rules’, of the ideal system being called ‘language’; and their artificial status may even be deemed to purge them from the ‘heterogeneous’ and ‘accidental’ qualities of ‘speech’ (cf. § 3). But in the account I have proposed here, the production of pseudo-data would ‘tune’ the language system out of its natural ‘frequencies’, and thus lead to the postulating of a pseudo-system which does not deserve to be called a ‘language’.

40. A second effect I would highlight has been the widespread textual strategy among linguists of formalising texts, apparently to render them more amenable to a scientific account (as in Koch, 1971; Ballmer, 1975). The actual result is to create a set of semi-texts that are artificial in a different way from pseudo-texts, namely in having some features or aspects of texts whilst many other features have been ‘abstracted away’ (§ 12, 18). These semi-texts are not trivial; in practice, they may be almost incomprehensible to everyone but their creators. The production of semi-data would thus also tune the language system out of its natural frequencies, and lead to the postulating of a semi-system which deserves to be called a ‘language’ no more than does a pseudo-system. The practices of the linguists during this production remain quite arbitrary and ad-hoc until we have a full and creditable account of which features or aspects should or should be ‘abstracted away’ and how the remainder should be ‘formalised’.

41. The practices of ‘formalising’ texts might correspond to a theory of the ‘text’ itself being a theoretical unit underlying a practical unit called ‘discourse’ (cf. van Dijk, 1972). Here, we might recall Hjelmslev’s aspiration to get the ‘text’ inside the same abstract theoretical sphere as Saussure’s ‘language’, which Hjelmslev himself never attempted to demonstrate in practice (§ 8). And, as I have remarked (§ 18), the demonstrations for ‘text grammars’ displayed an explosive complexity of rules and features, precisely insofar as a text is not solely a theoretical or virtual unit but also a practical and actual unit. If we disregard or ‘abstract away from’ its ‘practicality’ and ‘actuality’, we lose a host of significant constraints; trying to replace or reconstruct them all with rules and features is a gratuitous, self-defeating exercise.

42. Instead of complicating the relation between language and text with sets of pseudo-data and semi-data, the preferable alternative strategy would be to let the texts represent themselves (§ 82). The texts would retain the same representation as recorded speech or writing within the experience of the community of participants in text-events. Special annotations such as representing the intonation of speech in a written format should be clearly authorised by the purposes of the investigation and should not impede unduly comprehension or access to the data.

43. The practices of text linguists would then not entail transforming texts into some other representation but rather transforming our own modes of accessing and exploiting texts as data sources. We need to base our own authority not upon holding higher academic degrees in ‘theoretical linguistics’ (cf. §§ 4, 29) but upon examining large sets of authentic textual data (section H). 

C. Theory and practice 

44. These deliberations about the theories and practices of text linguists counsel us to move onto a higher plane (doing what Peter Hartmann used to call ‘Überhöhung’).⁵ There, we could define a ‘language’ to be a general theory of human knowledge and experience evolving in a dialectical relation to texts as a set of practices for working out the theory (cf.  Hartmann, 1963; Halliday, 1987). By that definition, all the members of a language community are implicated in the ‘theory and practice’ of language and text (cf. Beaugrande, 1997b, 1998c). More precisely, they sustain and share a dynamic theory which evolves through a criss-crossing interaction of many implicit and partial theories about how the practices of everyday life and ordinary talk are organised. Whilst the diverse contexts come and go, different partial sub-theories (rather than the whole theory) are applied and specified in and through the practices.

45. To be sure, a language is a highly unique type of theory. It cannot be effectively tested and verified or falsified in the familiar manner of a ‘scientific theory’, because it partially constitutes what it postulates. We cannot get outside language in order to talk about it without implicating ourselves in it (cf. § 38). Such would seem to be the aspiration of projects to ‘formalise’ language, but they merely end up replacing it a ‘semi-system’ whose relation to real language presents even more knotty problems (§ 40).

46. Nor again can we say that any one language is a ‘more correct theory’ or a ‘more valid theory’ than any other. The potential of any language for expressing human experience is infinite in theory, though always finite at any one moment of practice. Some languages have, for historical of political motives, had their potential more fully actualised in specific domains, such as science and technology. Such is true of English but by no means ‘validates’ the language to be the ‘superior’ one so often extolled in the discourse of ‘International English’ (discussion in Pennycook, 1994; Beaugrande, 1999b).

47. Now, by the definitions proposed above, what has been officially called a ‘theory of language’ or a ‘linguistic theory’ so far would be a ‘meta-theory’, whereas the texts we produce to formulate and expound the theory would display our ‘meta-practices’. According to my line of argument here, these ‘meta-domains’ would be fundamentally different depending on what their object of investigation is declared to be. In the meanings of Saussurian and Chomskyan linguistics, ‘language studied in and for itself’ is a theory about itself, about pure ‘theoreticalness’ disconnected from the practices of ‘actual speech’. In consequence, their ‘linguistic theory’ would be a ‘meta-meta-theory’, as shown most programmatically in the English title of Hjelmslev’s Prolegomena to a Theory of Language. This theory is doubly remote from practice, and the practices of ‘doing linguistics’ in general and ‘building linguistic theories’ in particular are radically under-constrained. Not surprisingly, many books on ‘theoretical linguistics’ have been sharply critical of the prior state of the field and have proposed to make a fresh start (Beaugrande, 1991: 334ff). Over time, ‘linguistic theory appears to offer a stunning variety and disparity of clashing doctrines’, and ‘striking divergences in terms, slogans, and technical contrivances’ (Jakobson, 1970: 12).

48. But if our object is declared to be authentic texts, then a ‘text-linguistic theory’ stays proximate to the practices of real language-users, and our own meta-practices would not be radically under-constrained. Indeed, we can productively invest our own status as participants in analysing or interpreting data produced for a community to which we rightly belong, such as the readers of the Times, so that our ‘text-linguistic practices’ are reconciled with the practices in our object domain, though of course not simply converging with them. Against the tendency of ‘scientific discourse’ ‘to hide itself’, as if it were somehow ‘not writing at all’ (Bazerman, 1988: 14) (§ 38), a ‘science of text’ would gain by displaying and even analysing its own discourse and the latter’s implication in the constitutive and constructive practices of text-events (see section H).

49. Working back toward ourselves from the opposite side, we should also devote far more attention to the ‘theoreticalness’ of producing and receiving texts in and about everyday life. Most producers and receivers would probably consider themselves ordinary practitioners or practical users of a language, who have no interest or authority concerning ‘theoretical’ matters. They might be irritated or perplexed to be told that knowing a language endows them with one among the most extensive and powerful theories a human can achieve, far beyond the revered but restricted ‘theories’ of ‘science’. But until they can acknowledge this vast endowment they will not be adequately empowered to use this resource for gaining access to knowledge and society (cf. Beaugrande, 1997a, 1998c).

50. In sum, a chief reciprocal task for a science of texts would be to cultivate an active sense of the ‘practicalness’ of scientific texts (ours in particular) and the ‘theoreticalness’ of ordinary texts, within a programme for reconciling the two sides within a dialectical interaction. Doing so would require reconstructing the ‘missing links’ between theory and practice, and between a language and the texts in that language. I shall now consider some prospects as we look ahead to the new millennium.  

D. Missing links

51. Perhaps the term ‘missing links’ sounds overly dramatic. After all, the linkage between a language and a text — between theory and practice — is achieved whenever human beings communicate, although they don’t recognise their achievement in those terms. But in terms of scientific investigation, our theories and models of this achievement undeniably manifest some ‘missing links’. I have been suggesting here that the toughest problems have stemmed from the assumption that a ‘language’ has a quite different mode of organisation than does ‘actual speech’ (or texts), and from the corresponding aspiration to describe language independently of actual speech. So the urgent move now would be to develop a more ‘text-like’ view of a language and a more ‘language-like’ view of a text (§ 55).

52. One promising approach would be the ‘systemic functional linguistics’ prominently developed by Michael Halliday, who was once a pupil of J.R. Firth. Halliday was among the first to realise that the order of a language must be ‘systemic’ in modes that anticipate, reflect, and support the ‘systemic’ uses of the language in texts (cf. § 21). Firth’s (1968: 192) vision of a text as a ‘longer piece’ to be ‘described as a relational network of structures and systems at clearly distinguished but congruent levels, converging again in renewal of connection with experience’ accorded with Halliday’s (1994: xiv) vision of ‘a language’ ‘as networks of interlocking options’. Quite plausibly, the networks constitute or contribute one type of ‘missing links’ whose precise nature is now beginning to be explained.

53. As Halliday (1997: 6) has recently acknowledged, his approach must contend with vast exponential size of the networks needed to represent ‘systemic potential’ as ‘alternative combinations of features’. The network needs to ‘present’ not just ‘those sets of options that are currently being instantiated’ but also the ‘open-ended’ ‘further expansion of that potential’ (1997: 7). Moreover, any such network representation should reflect the cline between options which are highly probable and options which are highly improbable (§ 36) (Halliday, 1993). Some options might be fully possible but rarely or never get used (§ 62).

54. So the networks for a language would be both too unwieldy and too under-determined to serve directly as networks for a text. In exchange, the networks for texts, such those I introduced for the pedestrian ‘rocket’ text (Beaugrande, 1980), are too compact and specific to represent the networks of the English language. Once more, we seem to detect some ‘missing links’, this time between two types of systemic networks.

55. The strategic place to look for ‘missing links’ would now be in a very large corpus of authentic texts. Such a corpus can offer our best means for promote both a text-like view of language, and a language-like view of texts (§ 51). It can also display the competence of the language community bent very far toward performance, as well as the community’s performance bent very far toward its competence (cf. § 12). We might undertake to complement the systemic functional approach by exploring appropriately sorted corpus data which may ‘provide evidence for our system networks, allowing them to extend much further in delicacy while continuing to model language as potential’ (cf. Halliday, 1997: 24).

56. Corpus data are so eminently suited to informing us about ‘networks’ because they offer concrete displays of the constraints upon how sets of choices can interact. In the ‘lexicon’ part of the ‘lexicogrammar’ of English, these constraints constitute the collocability in the virtual system, and the textual actualisations are the lexical collocations. In the ‘grammar’ part of the ‘lexicogrammar’, these constraints constitute the colligability in the virtual system, and the textual actualisations are the grammatical colligations. Following Halliday (1961) and Hasan (1986), we can say that the lexical choices are more delicate, and the grammatical choices are less delicate; ultimately, the lexicon would be the ‘most delicate grammar’, whilst the grammar would be the ‘least delicate lexicon’. Every point can access every other. Thus, what is ‘collocable’ can constrain not just the collocations but the colligations; what is ‘colligable’ can constrain not just the colligations but the collocations; and so on (cf. § 63). Such constraints ensure that the order of language is reasonably closed, whilst the order of texts is reasonably open (§ 26).

57. These prospects counsel us to adopt a synthetic approach of exploring how choices converge in texts, rather than the analytic approach of subdividing the text into steadily smaller pieces, as counselled by Hjelmslev (who never showed us how) (§ 8). We would then highlight the systemic continuity of the text event: not just the mere ‘co-presence’ of smallest meaningful units (e.g. ‘morphemes’), but the interaction of meanings within a coherence that is not just the sum of the meanings of the parts (Beaugrande, 1984).

58. Perhaps an analogy from particle physics might be helpful. There, the four ‘forces’ have recently been reinterpreted not as brute pushes and pulls, but as information exchanges of ‘messenger particles’: photons for the electromagnetic force, ‘gluons’ for the strong force holding the nucleus of the atom together, ‘bosons’ for the weak force regulating radioactive decay, and — presumed on grounds of consistency but not yet observed — ‘gravitons’ for gravity. By analogy, we could envision semantons as sub-symbolic messenger particles of meaning being interchanged among the virtual meanings of morphemes or whole words (cf. Smolensky, 1989; Beaugrande, 1997a). Through these ‘information interchanges’, some word-choices appear to ‘attract’ each other at varying degrees of strength, in analogy to magnetism or gravity, and thus to determine which areas of virtual meaning are being actualised to constitute the context (cf. § 63).

59. The interactive constraints and ‘attractions’ indicated in Fig. 1 would continually supply and sustain the links — ‘missing links’ insofar as they have eluded our investigations to date — between the language and the text. Part of the actualised output would be the cohesion and coherence of the text as ‘standards of textuality’, but also the other ‘standards’ identified in text linguistics: intentionality, acceptability, informativity, and situationality (Beaugrande and Dressler, 1981) (§ 15). The ‘most missing links’ of all begin to emerge, perhaps like the ‘cold dark matter’ astronomers believe to be holding galaxies together, namely the links of intertextuality. For the first time ever, we can assemble and compare the evidence for detailed and delicate intertextual constraints shared among hundreds or even thousands of related collocations or colligations. Instead of seeing the text as a set or sequence of sentences, we  can finally see it as a contribution to intertext (§ 35, 87).   

Click here to go to Text linguistics at the milennium 2