Logos and Languages

When I first made contact with text linguistics as a student in the mid-1970s, the field was in the process of gradually finding its identity. A convention had recently arisen of referring to previous work in linguistics as sentence linguistics. As far as I can discover, this term was nowhere adopted by the research in question, and still does not figure in language programmes. Whereas “text linguistics” currently appears in 1,161 websites on the Internet (10/2/2001), “sentence linguistics” appears in only 2, and one of them is a paper of my own (Beaugrande 1993).

No doubt many linguists would have judged the term a mere tautology, because they were firmly convinced that the sentence is the predominant unit of linguistic research. For Saussure (1916/1966, 124), “the sentence is the ideal type of syntagm”. For Sapir (1921, 35), “the sentence” “is the major functional unit of speech”. For Bloomfield (1933, 170), the “sentence” is the site where “linguistic forms” are “united” by “meaningful grammatical arrangement”. For Firth (1951/1957 170; 1956/1968, 148), “most linguists would agree that the study of the sentence as our primary datum is the order of the day”; and “the sentence” “naturally finds a central place” in a “general theory”. For Chomsky (1957, 13), a “language” is precisely “a set (finite or infinite) of sentences, each finite in length and constructed out of a finite set of elements”.

Philosophers of language have been even more expansive. Wittgenstein (1922/1963, 32, my translation) declared that “the totality of sentences is the language”. Wegener (1885, 181, my translation) opined that “all elements of language are originally sentences”. In a kindred spirit, Firth (1959/1968, 148) later speculated that “even the origins of all speech considered biographically in the nurture of the young and in the history of the race are to be found in sentences”.

Yet the consensus on the “central place” of the sentence was merely apparent, being deconstructed by the marked diversity among linguistic approaches. A careful examination of how the term was actually used in the discourses of linguists will detect a corresponding diversity of senses and meanings (Beaugrande 1999). A field which does not define its “primary datum” in a consistent and consensual way might seem to be a paradox. But the inconsistency has had the strategic value of enabling the field to shift its views of linguistic issues along a sequence of incisive steps while invoking the “sentence” as its guarantor of continuity and solidity.

For the purposes of the present paper, I shall highlight just three of these steps. In the first step, notwithstanding its already quoted status as “the ideal type of syntagm”, Saussure (1916/1966, 106), did not accept the “widely held theory making sentences the concrete units of language”, because the “totality of sentences that could be uttered” would reveal “immense diversity”; and “in no way do they resemble each other” (1916/1966, 106). So he proposed to consign it “to speaking [parole], not to language [langue]” (1916/1966, 106). But in fact he created a duality by consigning it to 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” (1916/1966, 124f). I hold this duality to be substantially correct for all language data, including the sentence; but Saussure failed to recognise it as powerful impediment for his famous dichotomy between langue and parole to be sustained for language data. So he inaugurated the tradition of double-tracking in linguists who announce dichotomies they find reassuring in theory yet recalcitrant in practice, and so proceed to circumvent them. He himself vowed to “deal only with linguistics of language”, whilst “using material belonging to speaking to illustrate a point” (1916/1966, 20).

[1] “Are we nearly there?” Alice managed to pant out at last. “Nearly there!” the Queen repeated. “Why, we passed it ten minutes ago! Faster!” (Through the Looking Glass)

[2] “You don't know how such a life tries a man like Kurtz,” cried Kurtz’s last disciple. “Well, and you?” I said. “I! I! I am a simple man.” (Lord Jim)

[3] you gave me reason to understand that you did care about him”. “Him! Never, never.” (Emma)

[4] he suddenly flung down his brush on the floor, said “Bother!” and “O blow!” and also “Hang spring-cleaning!” (Wind in the Willows)

[5] “My man's very angry.” “Phew!” said the bullocks. “He must be white!” “Of course he is,” said Vixen. “Huah! Ouach! Ugh!” (Jungle Book).

Bloomfield's definition also implies the corollary that we can’t tell what is in fact a sentence until we hear what is spoken. But he too double-tracked by demurring that “the linguist cannot wait indefinitely for the chance of hearing a given form used as a sentence” (1933, 179). And his own chapter on “sentence types” presents only inauthentic data, like ‘John ran away’, which do seem to have been invented rather than observed.

In practical fieldwork, analysts have often worked with “cleaned-up” text dictated sentence by sentence […] or with sentences repeated separately by the informant after having been given in consecutive text. Methodologically this is helpful; […] For theory, however, it can be fatal. One may, by limiting oneself to artificially selected data, build a theory of language structure which will not apply to language as it is in fact spoken.

Instead of “placing priority on dichotomous constructions”, Pike sought “theoretical advances” in an “emphasis upon unity”, and “abandoned the distinction between langue and parole proposed by de Saussure” (1967, 358, 536).

Our third step might resemble a dialectical antithesis to the other two. Far from seeing only “immense diversity” in the “totality of sentences that could be uttered” (Saussure), Chomsky (1957, 48, 54) proposed a “grammar” to “reconstruct formal relations among utterances in terms” of “structure” and to “generate exactly the grammatical sentences”. Indeed, his “transformational grammar” was founded squarely on the notion that the sentences of a language are not diverse but uniform — so much so that “rules” can be formulated to convert (derive, transform) any one sentence structure into any other.

The generative paradigm replaced Saussure’s dichotomy of langue and parole with a more mentalist one between “competence, the speaker-hearer’s knowledge of his language, and performance, the actual use of language in concrete situations” (Chomsky 1965, 4). A further dichotomy got applied to the sentence itself: “the central idea of transformational grammar” is that “deep and surface structures” are “in general, distinct” (1965, 16). This one too was reassuring in theory yet recalcitrant in practice, especially if “the grammar does not, in itself, provide any sensible procedure for finding the deep structure of a given sentence” (1965, 141). So Chomsky (1965, 18) circumvented his dichotomy between “deep structure” and “surface structure” by making the “simplifying and contrary-to-fact assumption that the underlying basic string is the sentence”.

The antithesis to Bloomfield’s definition of the sentence appears quite clearly in Lyons’ (1977, 29f) dichotomy between “system-sentences”, which are “abstract theoretical entities in the linguist’s model of the language-system”, and which “never occur as the products of ordinary language behaviour”; versus “text-sentences”, which are “utterances or parts of utterances, whether written or spoken”, and which “can be complete or incomplete”. Lyons noted that the “relation” between these two, “especially in casual or informal speech, may be quite complex” (1977, 30). Once again, the reassuring dichotomy proved recalcitrant and got circumvented: Lyons proposed to “operate with the simplifying assumption that system-sentences are sequences of words in a one-to-one correspondence with what would be judged, intuitively by native speakers, to be grammatically complete text-sentences” (1977, 30).

I have highlighted these three steps to point up how the priority allotted to the sentence in modern linguistics has encouraged strategies of double-tracking. The first step was to set it aside because it straddled the supposed boundary between language and speech, but to “illustrate” language with speech. The second was to define it as a unit only observed in speech, yet to describe language with data that had not been observed, but invented. And the third was to define it the very constitutive unit of language but to split it apart into “deep” and “surface”, and to rule out “observation” altogether. From there a dichotomy ensued between “system-sentences” and “text-sentences”.

This double-tracking was to prove hugely influential in motivating linguistics to look for language somewhere else besides observed speech. Saussure (1916/1966: 9, 11) already capitulated by asserting that “speech cannot be studied”, “for we cannot discover its unity”; it is merely a “heterogeneous mass of speech facts” — an echo of that “immense diversity” he attributed to sentences. Chomsky (1965: 4, 201) accentuated this posture by alleging that the “observed use of language” “surely cannot constitute the subject-matter of linguistics, if this is to be a serious discipline”; and that “much of the actual speech observed consists of fragments and deviant expressions of a variety of sorts”.

Such theoretical trends imply a remarkable but rarely noticed type of phrase transition or catastrophe (cf. Thom 1972): putting language into use abruptly moves from stable and integrative order over into unstable and disintegrative disorder. Why this should be so, and indeed how it could be so, I have never seen explained. But we can detect an implicit consensus that linguistics is best equipped to discover the “deep” or “underlying” order of language by extracting it out of its uses. This operation may be called abstraction, e.g.: “the study of competence abstracts away from the whole question” of “why speakers say what they say, how language is used in various social groups, how it is used in communication” (Dresher/Hornstein 1976, 328). Or, it may be called idealisation, which we are assured is “inevitable” (Lyons 1977, 586) and indeed “the sole means of proceeding rationally” (Chomsky 1977, 54). Under either label, the operation is nicely attuned to find in the sentence just what Saussure said: the “ideal type of syntagm” — seemingly detached, self-sufficient, well-formed, complete, and, above all, grammatical.

In common practice, the operation follows two major strategies, which produce two modes of second-order data:

(1) Linguists invent their own data that seem to stand by themselves, as in “the man hit the ball” (Chomsky 1957, 26ff). These I propose to call inauthentic data.

(2) Linguists replace natural-language data with an artificial notation, as in “a: ("x) [office (x) ®ƒ building (x)]; b: ("x) [building ®ƒ ($y) (has (x, y) & window (y))]” (van Dijk 1977, 100), which signifies that every office building has a window. These I propose to call non-natural data.

Often, both strategies are used: first you invent the data (‘the man hit the ball’) and then you convert them into the notation (‘T + N + Verb + NP’) (Chomsky 1957, 26f). Methodological objections can be forestalled by warning that “denying the admissibility of a constructed sentence” would be “a dangerously extreme view” (Brown/Yule 1984, 20).

I would assert the contrary: that these two strategies constitute the real dangers. Both of them place the linguists in the precarious role of creating and altering instances of language and thus raise an acute quandary regarding our authority. On what grounds can such data claim to represent the language as it can be observed in authentic and natural data? Saussure nowhere explained how linguists arrive at “langue” without contacting “parole”. When advising “the linguist” to “consider all other forms of expression”, and “not only correct speech” he seemed to forget his own warnings against the “diversity” which we saw to be his reason why “speech cannot be studied” (1916/1966: 6). Bloomfield also recommended “the linguist” to “observe all speech forms impartially”, but, as a devout behaviourist, asserted just the opposite of Saussure: “the linguist is in a fortunate position: in no other respect are the activities of a group as rigidly standardized as in the forms of language” (1933, 22, 37).

For Chomsky, a devout mentalist, the “universal practice” consisted of applying a “model that incorporates generative grammar” to the “actual data of linguistic performance” “along with introspective reports by the native speaker or the linguist who has learned the language” (1965, 18). His linguist was fortunate for a different reason than Bloomfield’s, namely in automatically commanding an “enormous mass of unquestionable data concerning the linguistic intuition of the native speaker, often himself”; in contrast, “sharpening the data by objective test is a matter of small importance” (1965, 20f). As such, linguists presumably hold the authority to act the role of the “ideal speaker-hearer in a completely homogeneous speech-community, who knows its language perfectly” (1965, 3).

Yet linguistics itself has been evasive about this authority. Saussure stipulated that “no complicated operation such as the grammarian’s conscious analysis is presumed on the part of the speaker”; yet “the sum of the conscious and methodological classifications made by the grammarian” “must coincide with the associations, conscious or not, that are set up in speaking” (1966, 167, 137f). Chomsky in turn denied that the “speaker of a language”, who has “mastered and internalised a generative grammar, is aware of the rules of the grammar or even” “can become aware of them”; and that “his statements about his intuitive knowledge are necessarily accurate”, since “a speaker’s reports and viewpoints about his behaviour and competence may be in error” (1965, 8). Apparently, linguists would constitute the exception, endowed by their academic training and status with super-human powers of introspection, at least where “grammar” is concerned.

I would suggest that these evasions are linked to the duality of the sentence Saussure recognised but could not appreciate. The sentence belongs neither wholly to language nor wholly to speech, so attempts to plant it on either side lead to evasions and circumventions. But attempts to plant it on both sides generate two disparate modes of sentences (or “structures”, “strings”, “sentoids”, and so on) whose mutual relation proves unworkably complex, and we make the “simplifying assumption” that they are the same after all.

The three steps I have cited in section 1 lead toward the view that a grammar of the sentence could claim adequacy and validity as a description of an entire language. This view attained the status of explicit programme in “generative linguistics”, one of whose adherents pronounced the magisterial aphorism in his inaugural lecture that “linguistics is not about language; it is about grammar” (Smith 1983, 4). The aphorism becomes even more portentous for a concept of “grammar” so abstract and remote from what the term “language” means to most people, including those who use their own language most skilfully and come the closest to being “ideal speakers”.

The same view was prevalent at the time when text linguistics was emerging. The new field seems to have begun in some quarters as an enterprise for resolving certain peripheral difficulties that had arisen in what was now being called “sentence linguistics”. Several authorities I interviewed told me as late as 1976 that the whole justification for text linguistics hinged upon finding phenomena which would belong to “grammar” in some established sense and which could not be explained on the basis of the single sentence (cf. Rieser 1977, 17).

Yet “text grammar” was typically conceived to be a counterpoint to “sentence grammar”, still comprising a set of “rules” and “features” suitably expanded to cross sentence boundaries. In the most direct correspondence, the “text” would be a “sentence sequence” and would be receive a grammatical description of its constituent sentences plus some mechanisms to link them (e.g., van Dijk/Ihwe/Petöfi/Rieser 1972). Still, the justification seemed to me stressfully problematic. We would risk confining text linguistics in a derivative and precarious status, living, albeit in a negative sense, in dependency upon sentence linguistics and exploring its margins and leftovers. We would have to react defensively whenever sentence linguists argued that “the properly linguistic facts of text can be taken care of within the existing framework of linguistic theory, and there is no reason to expect that a text-grammar will add anything” (Dascal/Margalit 1974, 82). Moreover, by “accepting certain sentence-grammatical preconditions and their methodological premises, we also procure their fundamental problems, such as the conception of semantics, the transformational apparatus, the status of syntactic and semantic features, the formulation of strategies for analysis” (Rieser 1977, 17, my translation). And, I would add, we procure the sentence, like a cumbersome piece of uninspected baggage out of which odds and ends are occasionally produced but whose full contents remain jumbled and uncatalogued.

In hindsight, the key issue appears to have been whether the text would undergo the same sort of career of antithetical steps as the sentence. Such might have already been implied when Hjelmslev (1943/1969, 17f, 39f) had counselled the “linguistic theoretician” to “provide the tools for describing or comprehending a given text” by “discovering certain properties present in all the objects that people agree to call languages”; “generalizing those properties and establishing them by definition”; and “setting up, for all objects of the nature premised in the definition, a general calculus, in which all conceivable cases are foreseen” — including even “a language without a text constructed in that language”. But he also suggested working in the opposite direction by “starting from the text as the datum” and “object of interest”, and producing “a self-consistent and exhaustive description through an analysis” (1943/1969, 12, 21, 16). Both procedures would be rendered obscure by the startling provision that “linguistic theory cannot be verified (confirmed or invalidated) by reference to any existing texts and languages” (1943/1969, 18). He didn’t explain, but his reasoning seems to have been that all texts are foreseen in the calculus a priori.

Moving up to the later line of argument advanced by Lyons for sentences, we would presumably get the “system-text” and the “text-text”. Though those terms are unlikely to be suggested, a similar dichotomy was proposed for “the term text to “denote the theoretical construct underlying the discourse” (van Dijk 1977, 3). One immediate and unsettling consequence, apparently to guarantee that the text be “well-formed and interpretable”, was to “disregard the possibility of dialogue” as a “sequence of utterances by different speakers”, whilst “assuming that such a sequence also may have the textual structure similar to monologue” (1977, 3) — yet another simplifying assumption, and one that creates obstacles for the observation of data like samples [1-5].

A “grammar” was proposed to be “a form-meaning-action rule system, in which abstract forms of utterances are related to both meaning and function of these forms in theoretically reconstructed contexts of communication” (van Dijk 1977, 2). As van Dijk, acknowledged, the three-part designation was intended to cover the familiar triad of syntax, semantics, and pragmatics. The lexicon figured only marginally as “specifying the meanings of expressions — words, morphemes, and phrases” (1977, 2).

Van Dijk’s grammar remained unofficially centred upon the sentence whilst providing also for the “composite sentence” and the “sentence sequence”. This strategy was probably encouraged by for his proposal to integrate “formal semantics” and “modal logics”, which led to a fresh dichotomy, this one between “modal sentences” and “non-modal sentences” (1977, 31). Not surprisingly, all of the data, except a few brief passages from a Mills-and-Boone type crime story and brief quotes from the Decameron and the Dubliners, were either non-authentic or non-natural, but all natural data were grammatical sentences in the ordinary sense. What the “abstract theoretical text” actually looks like — other than a sequence of sentence structures — I remain unable to determine in the work of any linguists who have advocated it (which van Dijk himself no longer does (cf. van Dijk/Kintsch 1983)

Gradually, text linguistics adopted a new course to establish that the text is a quite a different type of unit from the sentence, not merely grammatical, but cognitive and interactional. The focus of research accordingly shifted from text grammar to textuality, and at least some sources proposed cohesion as the general category for all issues otherwise subsumed under “grammar” or “syntax” (e.g. Beaugrande/Dressler 1981). And there the sentence was reduced to its ordinary status as the dominant grammatical pattern of careful written texts (cf. Halliday 1985, 192f).

The old dichotomies of langue and parole or competence and performance were discarded in favour of virtual system and actual system; in the long shadow of “virtual reality”, the term potential system seems preferable today (Beaugrande, in preparation). Far from ain implicit dichotomy between order and disorder in the tradition of Saussure and Chomsky, we have affirmed the principle that the two types of system must have respective modes of order whose designs support and guide each other within a dialectical cycle (Beaugrande 2000a).

Furthermore, we recognise intermediary subsystems like style, text type and discourse domain ensuring that the production and reception of texts do not involve the participants’ entire knowledge of the language (which Chomsky’s “competence” apparently would) (see now Kintsch 1998). These subsystems might be envisioned like high-level operating systems in a high-powered computer, where the interaction of many small constraints create large arrays of data. I would see here a linguistic analogy to the so-called Waltz effect, which I predicted some years ago (Beaugrande 1980, 294). Waltz (1975) had found just such an interaction among the physical constraints upon the shapes of visual objects reducing the number of possible interpretations for vertices from billions to thousands.

However, the factor of text types contributes a large problem space peculiar to the relation between language and text.. Aside from a few well-defined text types like “obituary”, “weather forecast”, or “house for sale”, the options of selection and combination still seem to imply some mode of “diversity” and “heterogeneity”, such as daunted Saussure. Moreover, types are frequently mixed. Sample [6] briskly mixes the discourses of solemnity (‘merriment, yesteryear’), social science (‘modernity, indigenous, identity, socio-economic development’), and tourism (‘age-old, timeless beauty and simplicity, life-styles’), along with the occasional solecism (‘much-privy glimpse’). The mix reflects multiple goals, such as disguising a tourist trap as a cultural site whilst flattering the readers’ command of an educated variety of English here in the Arab Gulf States.

[6] A wedding is a time for merriment and an apt occasion to showcase age-old traditions in an age where modernity is eroding important aspects of yesteryear. This much-privy glimpse of Arabia was a re-enacted wedding ceremony of the indigenous people, reflecting the timeless beauty and simplicity of Arabia's life-styles, customs and unique identity until the ’70s oil-boom brought in dramatic socio-economic development. (Khaleej Times)

But far from capitulating as Saussure did, we require that “speech must be studied”, and in sufficient quantities to substantiate the dialectical poles of uniformity and homogeneity.

This requirement is now much less unrealistic than in the times when my quoted sources appeared. Today, linguistics seems slated for a “scientific revolution” or “paradigm shift” under the impact of research using very large corpora (Beaugrande 2000). The reason seems plain: a very large sample of authentic data available for multiple query modes at long last utterly deconstructs any dichotomy between language and text, under whatever labels — langue and parole, competence and performance, deep structure and surface structure, and all the rest. What we see instead is an elaborately scaled continuity of rising delicacy between the selections and combinations that are more “language-like” and less delicate (e.g. those Word-Classes that can appear between an Article and its Noun in English) and those that are more “text-like” and more delicate (e.g. the referent of First Person Plural in Elizabethan history plays as compared to modern academic prose). Moreover, the corpus should, if properly analysed, help us to eventually determine what lies where on the scale. Provisionally, we can predict that the grammar is less delicate and lexicon is more delicate, but each of these has its own internal scaling of delicacy.

The corpus itself cannot be placed on either side. It is necessarily always far less than the language in two distinct senses: (1) on theoretical grounds, there will always be some potential texts that have not become actualised texts but would undeniably constitute authentic instances of the language; (2) on practical grounds, there will always be some actualised texts that don’t happed to be in the corpus; (2). Yet the corpus is also always far more than the text in two distinct senses: (1) on theoretical grounds, some texts show selections that might plausibly have been actualised in other texts of a comparable types but were not in that particular text; (2) on practical grounds, some texts were actively intended to be relevant to other texts, especially ones in the same conversation.

Our counterpart to the actual text cannot be the system-text as some “abstract theoretical entity”, which (to paraphrase Lyons) exists only “in the linguist’s model of the language-system” and which “never occurs as the products of ordinary language behaviour”. We have defined the text as a communicative event, so that its status depends upon its occurrence and not upon the theoretical specifications of a text grammar. The question of why some texts do not occur is not a grammatical one, and cannot be productively pursued until we have examined a wide range of the texts that do occur.

Our central concept is rather the intertext, that is, a set of comparable texts that are all authentic and natural by virtue of their attested occurrence. The concept must be very wide and flexible because the range of intertexts is. So too is the concept of intertextuality, which was still barely explored in early text linguistics (cf. Beaugrande/Dressler 1981), is a key term today in text linguistics as well as such fields as semiotics, literature theory. (The Internet returned 6,636 web pages for the term on 10.2.2001.)

But the widespread use of a term need not indicate a consensus about its meaning, as we saw in section 1 for the “sentence”. Obviously, the texts constituting an intertext must share resources of the language and especially of the grammar, but how and how far they do so cannot be determined by the modes of top-down abstract theoretical argumentation that have surrounded the sentence. We shall need massive bottom-up searches and analyses of data sets and a wholly open mind about the means and results of our observations. And we must oppose any dichotomies that impede the description of authentic data and oblige us to circumvent them by means of “simplifying assumptions” like those I have cited. Our own simplifying assumption is that the set of authentic texts we can encode and submit to a concordance program represents a reasonable approximation of the intertext, since we cannot prove this. And this assumption animates rather than impedes us in describing authentic data. Instead of representing the text by means of some symbolic or logical formalisation like the one I quoted for having windows in every office building, which entrains us in producing non-natural data, we let the text represent the text, and our accessible set of texts represent the intertext, which allows us to invest our competence as ordinary producers and receivers of natural data (Beaugrande 2000).

For the reasons outlined in section 2, the interest in “text grammar” went into eclipse when textuality came to the fore; now, intertextuality might seem capable of bringing it back – like some Hegelian dialectic of thesis, antithesis, and synthesis. But the picture is not so straightforward. The dominant types of “sentence grammar” and “text grammar” that preceded the rise of textuality were formalist and theory-driven; the text grammar that seems poised for a new career is functionalist and data-driven. Halliday’s (1994, xvi) pronouncement has come to its full force: a “text linguistics” or “discourse analysis” “carried on without grammar — or even as an alternative to grammar” is “not an analysis at all, but simply a running commentary on a text”. We need a “comprehensive view of grammar” for “interpreting a text in its context of culture” (1994, xxxii).

The upcoming “paradigm shift” can be predicted to transform the entire concept of a language: not a static (synchronic) system of units (e.g., phonemes, morphemes, phrases, sentences) but a dynamic system of relations. Instead of the old dichotomy between “langue and parole” or “competence and performance”, we can recognise a dialectical cycle between combinability (language as potential system) and combination (text as actual system). And instead of the dichotomy between a grammar of rules and a lexicon of expressions, we can explore the unified lexicogrammar to describe the typical grammatical combinations called colligations and the lexical combinations called collocations, and thus turn our attention from what is or is not grammatical to what is or is not colligable or collocable.

These conceptions might help us to distinguish between authentic data and non-authentic data. Here are three favourite samples from linguistic discussions intended to illustrate the concept “grammatical sentence of English”:

In July 1994, I submitted them and their main collocations to the Bank of English (BoE) at the University of Birmingham, the world’s largest corpus, then at 224 million words of authentic text. None of them was attested, not because they are ungrammatical, or inadmissible on any other grounds (unacceptable, incoherent etc.), but because they are non-authentic: detached, encapsulated trite, vacuous. The only combination of “man + hit + ball” was [10], probably spoken by parent to child. For “eager to please”, three instances were attested [11-13], each with a Direct Object for “please” that was missing in [8] and with more interesting Subjects than our old friend “John”. For “cats sitting on mats”, the only attestations were derived from the use of this example in schoolbooks or logician’s debates, e.g. [14-15], unrelated to any real cat. These are authentic derivations from non-authentic data and appeared in discourse purporting to explain language in the abstract.

[14] On the first page was a drawing of a brindled cat seated on a recognisable mat, the original “cat on the mat” now quoted in derision of an antiquated method of teaching

[15] material-objects statements, “There is a cat on the mat”, statements about people in novels, statements of mathematics

A text grammar centred on a corpus would not invent data at all, because, contrary to the assertions quoted above, the “linguistic intuition” of the linguist by no means yields “unquestionable data”. Intuition is plausible and opportunistic: it can reliably assess and interpret authentic data place before us; but it cannot reliably produce authentic data. Nor have we any motive to do so, given the huge quantities of authentic data now at our fingertips (or mouse buttons).

Colligability would be developed by scanning the preferences of selection and combination among the totality of options allowed the grammar of a language. For example, the Collins COBUILD English Grammar (based on a BoE corpus of 20 million words) says that “using the appropriate Adjective preceded by ‘the’ to “talk about groups of people who share the same characteristics” “is possible for almost any Adjective” (Sinclair et al. 1991: 21). When the BoE reached 336 million words, Sinclair (1998: 81) reported that “the unfortunate are very definitely the recipients of this structure”, such as ‘the elderly’, ‘the injured’, ‘the unemployed’, ‘the sick’, ‘the aged’, ‘the worthless’, ‘the poor’, and ‘the mentally handicapped’”; “‘we do find antonyms of some of these: ‘rich’ as well as ‘poor’, ‘young’ as well as ‘old’, ‘educated’ as well as ‘uneducated’, but their role is to occur with the unfortunate people for contrast; they are not heavily used on their own”.

In my own 15-million-word corpus of British and American writers dating roughly between 1750 and 1920, I found contrasts like those noted by Sinclair, e.g. [16-17], but also many where the fortunate people occurred alone, although sometimes with the ironic twist of not being secure in their good fortune [18-19].

[16] Smile with the simple and feed with the poor? […]; let me smile with the wise, and feed with the rich (Life of Samuel Johnson)

[17] None know the unfortunate, and the fortunate do not know themselves (Poor Richard’s Almanack)

[18] There is always some levelling circumstance that puts down the overbearing, the strong, the rich, the fortunate, substantially on the same ground with all others (Emerson’s Essays)

[19] the educated see a menace in his [the black man’s] upward development (Black Souls)

Here we are probably seeing the expanded creativity and stylistic awareness of literary discourse.

For example, some Verbs are mostly used only in the Active2 (e.g. ‘elude’) or only in the Passive (e.g. ‘construe’) (according to the Collins COBUILD English Language Dictionary 458, 302). We do not claim that using such Verbs the other way round is ungrammatical, but that it is improbable. In my own corpus of British and American writers, I found only 2 out of the 74 uses of ‘elude’ to be in the Passive:

[20] they lessen the consumption; the collection is eluded; and the product to the treasury is not so great (Alexander Hamilton)

But 21 out of 59 uses of ‘construe’ were found in the Active, not just in the current sense of ‘interpret’ [22] but also the senses of ‘translate’ [23] or ‘interpret something into something else’ [24].

[22] This behaviour in her niece the good lady construed to be an absolute breach (Tom Jones)

[23] he recalled the shrewd northern face of the rector who had taught him to construe the Metamorphoses of Ovid in a courtly English (Portrait of the Artist)

[24] She’s an excitable, nervous person: she construed her dream into an apparition (Jane Eyre)

To expand from colligability to collocability, we can examine the ways in which a Verb prefers certain classes of Subjects or Objects. The Verb ‘warrant’ was attested in the Boe (thne at 226 million words) in 392 data lines; to make the sample precise, I hand-sorted a selection of 228 lines by eliminating repetitions, e.g., when a statement by a politician got reported in several media; and false alarms where the software had inadvertently return the key word as a Noun. Fully 224 lines had a subject in the Third Person, just 4 in First Person, and none at all in the Second Person; and, within the Third Person, I found a mere handful of Pronoun Subjects: ‘he’ (6 lines), ‘she’ (0), ‘they’ (5), and ‘it’ (7). These data showed that the Verb is almost never used for people except in a special sense, namely when you say what you believe though you don’t have evidence, as in:

[25] The soil may look innocuous enough when you’ve dug it over but I’ll warrant it’s teeming with root-eating wireworms. (BoE)

Otherwise, the Verb is chiefly found in contexts of culture when some event or state, carrying a Pejorative Value, is said to ‘warrant’ or ‘not warrant’ some reaction, as in:

[27] the disability is not felt sufficient to warrant his inclusion in the wheelchair (BoE)

Many of the Subjects and Objects fell into associative classes that are not unduly hard to label, e g.:

(a) subjects: actions: ‘achievement, aggressions, behaviour, brawl’; resources: ‘abilities, acreage, growing area, scrappable cars’; knowledge: ‘evidence, information, perception, scientific authority’; messages: ‘message, complaints, accusations, revelations’; problems: ‘impropriety, violence, casualty rate, food shortage, ill health, disability, job bias, discriminatory practices, slowing in the economy, degenerating trees’;

(b) objects: (in)appropriate reactions: ‘further action, change, commitment, conclusion, consideration, expansion, extension, formation, increases, plan, step, treatment’; consumption of resources: ‘cost, expenditure, loss of any troops’ lives, overeating, steeper taxes, shelf-space’; messages: ‘comment, phrase, suggestion, description, mention, apology, appellation, billing, briefing, brochure, serious talk, talking-to’; knowledge-gathering: ‘attention, consultations, examination, retrospective survey, review, hearing, inquiry, investigation, trial’; solving problems: ‘economic assistance, intensive care, sending of those supplies, using these drugs, making peace, breaking the embargo, easing of interest rates’; retaliating: ‘retribution, penalties, charges, jail time, lengthy ban, massive American retaliation, pre-emptive strike, criminal prosecution, trial, capital punishment’.

[27] these old homes are chilly enough to warrant guests wearing thermal long johns (BoE) (problem - problem solving)

[28] encouraging progress did not yet warrant full-scale economic assistance (BoE) (resources - consumption of resources)

I found a number of cross-overs, such as an action as Subject and a message about it as Direct Object as in [29]; or a message Subject and a reaction as Object [30]

[29] the operation was important enough to warrant a middle-of-the-night briefing (BoE

A “grammar” for “interpreting a text in its context of culture” (Halliday) would also note that declaring what does or does not ‘warrant’ what implies that the event or state of affairs is in some way unusual or significant enough that a reaction is in order, and that those who might be expected to react should say why or why not, and how. Accordingly, the speaker or writer — or, when the discourse is reported, the originator of the message — purports to represent some institution or authority, and my data suggested what kind: government, judiciary, military, sports, business, science, and medicine. Or, if the person does not, then the use of ‘warrant’ implies a subtle signal that authority is being claimed anyhow; as when a journalist said”

My own corpus of British and American writers showed different results. Among 210 attestations, 80 occurred with a First Person Subject. The colligations ‘I warrant’ and ‘I’ll warrant’ were always deployed to assert what one believes and doesn’t have evidence, as in [32-33]. But in other colligations I found the older meanings of ‘certify’ [34] and ‘guarantee’ [35-36], the latter being etymologically related as well, and attested in Early Modern English [37].

[32] I warrant you will never repent having the money into his hands. (Tom Jones)

[33] ‘But where shall I go if you drive me away? What shall I do?’ ‘Oh, I’ll warrant you know where to go and what to do.” (Jane Eyre)

[35] I think that I may warrant you one worm to every three sods you turn up, if you look well in among the roots of the grass (Walden)

[36] the slabs are the best of their kind, and I can warrant them to resist rain and frost for a hundred years (Far from the Madding Crowd)

[37] This Ryche man thenne sold his oylle to the marchaunts and waraunted eche tonne al ful (William Caxton, The subtyl historyes and fables of Esope, Auyan, Alfonce, and Poge, 1484)

Here, we can be fairly certain we are confronting the historical dimensions of authentic data.

Much additional data could be presented here; in fact, the data are so plentiful that any such text grammar would demand the collaboration of a large team of linguists. Here, I can only sum up the prospects and relate them to issues and problems raised in this paper. The major difference between a language grammar and a text grammar can be seen in its concern for colligability and collocability. A language grammar needs to cover the totality of options including ones that are extremely rare in authentic data, such as ‘will have been going to have been taking’ (Halliday 1985, 181). Even the Future Perfect Progressive (‘will have been’ + Present Participle) is highly uncommon, not appearing once in my own corpus. A text grammar can regard this totality like a backdrop against which we can project some selections and combination in low relief, and others in high relief. We may find the grammatical structure illustrated by ‘the man hit the ball’ to merit high relief; but we cannot just assume it does without confirmation by data, nor present it as some supremely valid or canonical sentence-type, as much work in formal grammar has done (e.g. Chomsky 1957).

I am not suggesting that we can actually extract our text grammar from a particular text, as was paradoxically attempted for a formal text grammar (van Dijk,/Ihwe/ Petöfi/Rieser 1972). But with sustained and strategic work, we can, I suggest, extract it from an intertext. In some data, our work will be near or at the surface, namely at high degrees of delicacy where sets of choices are nearly fixed, as in ‘you’d be surprised at the number of people who…’ reported from the BoE by Sinclair (1998, 83). But far more often, our work will concern interactive word-classes of the type briefly illustrated from the Subjects and Objects of a Verb like ‘warrant’. And those classes will certainly reflect many rich contexts of culture, as stipulated by Halliday.

Our descriptive terms should be no more technical than is necessary to do justice to the data. We cannot know in the early stages, let alone in advance, which terms will prove most applicable. Some terms in both traditional and formal grammar will doubtless be needed, such as “Noun” and “Verb”, but, since those are situated in language grammar, they must be extensively qualified and specified for a text grammar (as set forth in Beaugrande 1997). And the specifications will have to take account of form, function and meaning as these three factors can be demonstrated in authentic data sets (Beaugrande 2001).

Non-technical terms are also required so that our findings can be shared with, and judged by, the widest possible cross-section of language users comparable to those involved in our corpus data — not just linguists who are fluent in technical terms, but authors, public speakers, journalists, media personalities, political leaders, and so on. They would be welcome contributors to balance out our own limitations and biases, especially when we must rely on intuition.

Furthermore, the term “text grammar” itself is just a non-technical substitute for the more accurate but cumbersome term intertext lexicogrammar, as I shall briefly explain. A single text cannot have a grammar of its own, for it could then not be used to communicate. A text can only manifest a set of selections and combinations from the grammar of its intertext. Here, we partly agree with the early text grammars in seeing grammar as more abstract than the text; but we disagree sharply in assigning the grammar to a large corpus of authentic texts as an intertext. The intertext is open, as I have pointed out, and also cannot have a grammar all its own, but the larger and more balanced it becomes, the higher the probability of being representative for the grammar of the language. The ultimate stage of equivalence cannot be reached any more than any one speaker can “know the language perfectly” (the Chomskyan ideal). But the grammar of a very large corpus can approximate the grammatical knowledge of those speakers and writers who produced the text, and of those hearers and readers who received them (or were intended to) — and also the grammatical knowledge of the linguists examining the corpus data. Instead of claiming complete and definitive authority by virtue of an academic training and status that instil super-human powers of introspection (as the generative approach seems to imply), we at most claim limited and tentative authority by virtue of extensive experience in describing large data sets. And instead of situating the abstract texts in “theoretically reconstructed contexts of communication” (van Dijk), we situate them in empirically demonstrated real contexts.

Moreover, we must have a lexicogrammar because corpus data also deconstruct the long-standing dichotomy, also going back to Chomsky, Saussure and even to Henry Sweet, between grammar as the domain of order and lexicon as the domain of disorder. Demarcating the borderline between grammar and lexicon is simply not workable in an open-minded description of corpus data, for the compelling reason that the one provides rich constraints upon the other, not just occasionally but systematically (Francis 1993). To separate them would be to reject unassessed a vast store of information both about the language and about the text.

The relation between the intertext and the whole language at higher generality and also the relation between the intertext and the individual text at lower generality, are undoubtedly complex, but the complexity is of a different order than in Lyons’s relation between “system-sentences” and “text-sentences” cited in section 1. When we explore corpus data, we are dealing with natural complexity, which is managed in discourse processing by the interaction of parallel modes of constraints (Rumelhart, McClelland et al. 1986; Kintsch 1998); and this same processing supports the analysis performed by linguists in their role as members of the discourse community. When we postulate the text to be an abstract theoretical construct underlying the discourse (van Dijk 1977), we are dealing with non-natural complexity, which incurs the inevitable loss of the natural complexity of authentic discourse, as I have briefly demonstrated.

These prospects clearly indicate why a new career of text grammar must be more carefully defined and qualified than was done in its previous career. Undoubtedly, our conceptions about “grammar” and due for some incisive revisions which will bring it much closer to the text than was true in the past. We will then be far better positioned to apply the grammar to real texts for a range of purposes, among which the teaching and learning of language in general and of grammar in particular will surely stand out.

Beaugrande, Robert de (1980), Text, Discourse, and Process: Toward a Multidisciplinary Science of Texts, Norwood, N.J.: Ablex.

– (1993), “Discourse analysis and literary theory: Closing the gap between linguistics and literary study”, in: Journal of Advanced Composition 13/2, 423-448.

– (1997), New Foundations for a Science of Text and Discourse, Greenwood, CT: Ablex.

– (1999), “Sentence first, verdict afterwards: On the remarkable career of the “sentence”, in: WORD 50, 1-31.

– (2000), “Text linguistics at the millennium: Corpus data and missing links”, in: Text 20.

– (In preparation), A New Introduction to the Study of Text and Discourse.

Beaugrande Robert de/Dressler, Wolfgang (1981), Introduction to Text Linguistics, London: Longman.

Brown, Gillian/Yule, George (1984), Discourse Analysis, Cambridge: Cambridge UP.

– (1977), Language and Responsibility (with Mitsou Ronat), New York: Pantheon.

Dascal, Macelo/Margalit, Avishai (1974), “Text grammars: A critical view”, in: Ihwe, Jens/ Rieser, Hannes/Petöfi, János (eds.), Probleme und Perspektiven der neueren textgrammatischen Forschung 1, Hamburg: Buske, 81-120.

– (1977), Text and Context: Explorations in the Semantics and Pragmatics of Discourse. London: Longman.

Dijk, Teun van/Ihwe, Jens/Petöfi, János/Rieser, Hannes (1972), Zur Bestimmung narrativer Strukturen auf der Grundlage von Textgrammatiken. Hamburg: Buske.

– /Kintsch, Walter (1983), Strategies of Discourse Comprehension, New York: Academic.

Dresher, Elan/Hornstein, Norbert (1976), “On some supposed contributions of artificial intelligence to the scientific study of language”, in: Cognition 4, 321-398.

Firth, John Rupert (1957), Papers in linguistics 1934-1951, London: Oxford UP.

– (1968), Selected papers of J.R. Firth 1952-1959 (ed. Frank R. Palmer), London: Longman.

Francis, Gillian (1993), “A corpus-driven approach to grammar”, in: Baker, Mona/ Francis, Gillian/Tognini-Bonelli, Elena (eds.), Text and Technology: In Honour of John Sinclair, Amsterdam: Benjamins, 137-156.

Halliday, Michael (1967), Intonation and Grammar in British English, The Hague: Mouton.

– (1994), Introduction to Functional Grammar, Second revised edition, London: Arnold.

Kintsch, Walter (1998), Comprehension: A Paradigm for Cognition, Cambridge: Cambridge University Press.

Pike, Kenneth (1945-64/1967), Language in Relation to a Unified Theory of the Structure of Human Behavior, The Hague: Mouton.

Rieser, Hannes (1977), Textgrammatik, Schulbuchanalyse, Lexicon, Hamburg: Buske.

Rumelhart, David/McClelland, James et al. (1986), Parallel Distributed Processing: Explorations in the Microstructures of Cognition (2 vols.), Cambridge, MA: MIT Press.

Saussure, Ferdinand de (1916/1916), Course in General Linguistics (transl. Wade Baskin), New York: McGraw-Hill.

Sinclair, John (1998), “Large corpus research and foreign language teaching”, in: Beaugrande, Robert de/Grosman, Meta/Seidlhofer, Barbara (eds.), Language Policy and Language Education in Emerging Nations: Focus on Slovenia and Croatia, Stamford, CT: Ablex, 79-86.

Smith Neil (1983), Speculative Linguistics: An Inaugural Lecture, London: University College.

Thom, René (1972), Structural stability and morphogenesis (trans. D.H. Fowler), New York: Addison-Wesley.

Waltz, David. (1975), “Understanding line drawings of scenes with shadows”, in: Winston, Patrick (ed.), The Psychology of Computer Vision, New York: McGraw-Hill, 19-91.

Wegener, Philipp (1885), Untersuchungen über die Grundlagen des Sprachlebens, Halle: Niemeyer.

Wittgenstein, Ludwig. (1922/1963), Tractatus Logico-Philosophicus, Frankfurt: Suhrkamp.

1 Today, we can recognise this definition to be for the Tone Group (Halliday 1967, 1994), but Bloomfield’s conception of language in terms of segments and constituents led him to the unworkable notion that intonational phenomena like “pitch” were to be described in terms of “secondary phonemes” (1933, 107f).

2 I capitalise grammatical terms to prevent confusion with ordinary terms.

Text Grammar Revisited

Robert de Beaugrande