1. We suggested in III.14 that the stability of the text as a system is upheld via a continuity of occurrences. The notion of “continuity” as employed here is based on the supposition that the various occurrences in the text and its situation of utilization are related to each other, or in cognitive terms: each occurrence is instrumental in accessing at least some other occurrences. The most obvious illustration is the language system of syntax that imposes organizational patterns upon the surface text (the presented configuration of words). In using the term “cohesion” (“sticking together”), we wish to emphasize this function of syntax in communication.1
2. The human mind is rather limited in its capacity to store surface materials long enough to work on them (cf. Keele 1973; Loftus & Loftus 1976). Materials are placed in active storage, a “working memory” where processing resources are distributed among elements of a presentation according to their importance (cf. Eisenstadt & Kareev 1975: 338f.; III.2, 6; V.4, 10). There appears to be a very short-lived imprint of visually or acoustically perceived materials upon which some provisional organization must be rapidly imposed (cf. Sperling 1960; Neisser 1967; Crowder & Morton 1969; Rumelhart 1970). These provisionally organized materials can then be retained for some longer periods of time, but still only within modest limits. It follows that text processing could not afford to run through the participants’ vast stores of world knowledge immediately; there must be some ancillary organizational system with far more limited options and patterns. In natural language texts, this system is that of syntax, whose classes and structures, though often more diversified than what is found in English, are still quite limited in number in comparison to the classes and structures for concepts and relations (cf. III.25ff.; V.30). This account is borne out by observations that surface structures are more predominantly maintained in a “short-term” storage, and conceptual content in a “long-term” storage (Wright 1968).
3. The functions of syntax reflect these cognitive factors. Since grammatical dependencies often obtain among elements not directly adjacent to each other (III.26), syntax must provide closely-knit patterns of various size and complexity into which current materials can be fit.2 Hence, the major units of syntax are patterns of well-marked dependencies: the phrase (a head with at least one dependent element), the clause (a unit with at least one noun or noun-phrase and an agreeing verb or verb-phrase), and the sentence (a bounded unit with at least one non-dependent clause).3 These units can all be utilized in a short span of time and processing resources. For long-range stretches of text, there are devices for showing how already used structures and patterns can be re-used, modified, or compacted. These devices contribute to stability (cf. III.14) and economy (cf. V.15) in respect to both materials and processing effort. Recurrence is the straightforward repetition of elements or patterns, while partial recurrence is the shifting of already used elements to different classes (e.g. from noun to verb). Repeating a structure but filling it with new elements constitutes parallelism. Repeating content but conveying it with different expressions constitutes paraphrase. Replacing content-carrying elements with short placeholders of no independent content constitutes the use of pro-forms. Repeating a structure and its content but omitting some of the surface expressions constitutes ellipsis. One can also insert surface signals for the relationships among events or situations in a textual world, namely by using tense, aspect, and junction. The ordering of expressions to show the importance or newness of their content yields functional sentence perspective. In spoken texts, intonation can also signal importance or newness of content.
4. Cohesion within a phrase, clause, or sentence is more direct and obvious than cohesion among two or more such units. Even so, the issue of how these closely-knit units are built during the actual use of a text is worth our consideration. Procedurally, the basic phrases and clauses of English can be viewed as configurations of links between pairs of elements, many of them having further linkage (cf. Perlmutter & Postal 1978; Johnson & Postal 1980). The question is then: how and in what order are these links created?
5. Abstract grammars suggest various answers to the question, but in general, the real-time processes involved have not been prominent criteria for setting up such grammars. We would like to point out a different kind of syntax which has been demonstrated to perform the best in the simulation of language processing on computers: the augmented transition network (cf. Thorne, Bratley & Dewar 1968; D. Bobrow & Fraser 1969; Woods 1970; Christaller & Metzing (eds.) 1979). The network is a configuration of nodes, in this case, grammar states,4 connected by links, in this case, grammatical dependencies. To move from one node to another, the processor performs a transition across a link. This operation demands the identification of the link as one of a repertory of dependency types, e.g. “subject-to-verb” or “modifier-to-head”. The transition can be augmented with any kind of search or access operation, such as specifying the exact category to which the upcoming node belongs (Winston 1977: 172). A special kind of augmenting could test what conceptual relation corresponds to the grammatical dependency being created (cf. V. 30).
In a transition network, the structures of phrases and clauses are
operationalised as means to build and test hypotheses about the types of
elements to use or expect at any given time. Hence, these networks capture the
grammatical strategies and expectations
of language users; and they express the rules of grammar as procedures
for using the rules (Rumelhart 1977a: 122). The phrase, clause, or sentence
appears as an actually occurring grammatical macro-state
in which elements are micro-states
of the textual system.5 The divergence between competence and
performance is reconciled (vis-à-vis Chomsky 1965), because the rules are
intended to stipulate the actual, rather than
the virtual roles of grammatical
dependencies (on actual/virtual, cf. III.1 2).6
7. We can provide here only a brief glimpse of the transition network in action (for more details, see Winston 1977; Rumelhart 1977a; Beaugrande 1980a, b). Let us model the processing of this simplified version of the opening of sample  in I.1:7
[4.1] A great black and yellow rocket stood in a desert.
As we noted in III.26, the linear sequence is partly misleading, since the several modifiers are at unequal distances from their head ‘rocket’. Therefore, one main operation will be to maintain the “modifier-to-head”8 dependencies as direct links. As soon as the determiner ‘a’ is set up, the processor enters a noun-phrase network, i.e. a macro-state, in which a head noun has at least one element depending on it. The processor sets up the goal of accessing the head; thereupon, the head will be used as a control centre for the whole macro-state.9
8. Figure 1 shows the processor moving along through the noun-phrase network. It keeps predicting the head, but finding modifiers instead. Presumably, the hypothesis of head is given preference, but the hypothesis of modifier is next in line to be tried (the order of preferences being of course subject to variation in different languages—we are only using English for our demonstration). We show this operation with a dotted line link for the failed hypothesis, and a continuous link for the successful one. When the junctive ‘and’ occurs,10 the processor can confidently predict that (a) another modifier is forthcoming, and (b) this will be the last modifier before the head. These predictions are confirmed, so that the head is attained and links between it and all its dependent elements can be filled in as shown in the upper part of Fig.1. We can look at the operations in another perspective. The processor would place each occurrence of a single element on a hold stack until the entire macro-state was constructed, and then would build a grammatical dependency network from the results. A “hold stack” can be used as a “pushdown” storage where elements are entered in a certain order and removed in the reverse order. Figure 2 illustrates the order for stacking our noun phrase: the times of entry are on the far left, the labels of states next to the times, and the states themselves in the middle. When the head is found, the processor creates the network structure to the
right. Presumably, the elements would be taken off the stack in the reverse order from that of their entry. The small numbers along the lines are intended to indicate the order of the linkage according to that principle. However, the procedures actually used during communication may be more variable.11
9. The rest of our sample would be run as the construction of a verb-phrase network. This macro-state is entered upon encountering the verb ‘stood’, which is already the head. The processor can safely anticipate a modifier, but it would be expedient to augment the search by shifting to a greater degree of detail and looking for subclasses of modifiers, e.g. adverb vs. prepositional phrase. If the adverb were the preference hypothesis, it would fail and yield to the prepositional phrase, the latter being a macro-state within the overall verb-phrase macro-state. The sub-goal is set up of finding the head of the phrase, which, in this sample, follows its determiner (‘a’) immediately (‘desert’). Figure 3 shows the parsing out of the verb-phase in terms of a system of states, like Fig.1.
10. We end up with the sentence shown not as a linear sequence, but as a labelled transition network. The nodes are the grammatical states and the links are the dependencies. Figure 4 illustrates the network in this fashion.
The role of such a network would be to organize the surface structure according to the most direct access, so that the linear text could be read off it during production, or traced back to it during reception.12 Although research on production is still very rare, there is some empirical evidence on reception that supports the model of the transition network. Stevens and Rumelhart (1975) found that people’s syntactic predictions about how sentences would continue beyond a particular point agreed some 75 per cent of the time; and when readers altered a text while reading aloud,13 their changes agreed 80 per cent of the time with their expectations established in the other tests. These proportions of agreement are strikingly high and should be sufficient for workable processing. The application of expectations to the actual input would entail only minor specification and modification as foreseen by the notion of procedural attachment (cf. III.19). In terms of networks, the transitions could be suitably augmented (cf. IV.9). The expected patterns would yield a reasonable match with current materials most of the time.
11. In closely-knit units such as phrases, clauses, and sentences, cohesion is upheld by fitting elements into short-range grammatical dependencies. In long-range stretches of text, the major operation is discovering how already used elements and patterns can be re-used, modified, or compacted. The devices enumerated in IV.3 fulfil that function via repetition, substitution, omission, and signalling relationships. These devices are far less obligatory than those which serve for closely-knit units: within the latter, missing elements are more noticeable and disturbing in immediately active storage.14 Failure to complete a clause or sentence would be more disorienting than failure to use recurrence, pro-forms, junctives, and so on. The long-range devices are thus contributors to efficiency rather than being grammatical obligations (Beaugrande 1980a): they render the utilization of the surface text stable and economic (IV.3).
12. The direct repetition of elements can be called recurrence, since the original occurrence merely happens again (cf. Plett 1975). Recurrence appears on various levels. Weinrich (1972) shows that grammatical categories tend to recur rather than shift—a finding obtained by Harris (1952) via a different approach (cf. II.21f.). Van Dijk (1969) suggests that components of concepts recur to support the coherence of texts. But we will glance here only at lexical recurrence, that is, repetition of the same words or expressions, as being the most noticeable sort.15
13. Recurrence is common in spontaneous speaking, where restatement results from short planning time and rapid loss of the surface text. Following a flash flood, a distraught county supervisor made this observation (reported in the Gainesville Sun, 20 Dec. 1978. We adopt the convention of underlining demonstration elements.):
 There’s water through many homes I would say almost all of them have water in them. It’s just completely under water.
When there are more resources and time available for text production, recurrence is customarily kept within limits. If unduly frequent, it lowers informativity (in the sense of I.17f ) For this motive, Georgia Green (1968: 22) suggests that an utterance like:
 John ran home and John ran home.
would be unacceptable, since it is pointless to say exactly the same thing twice. However, recurrence is prominently used to assert and re-affirm one’s viewpoint,16 or to convey surprise at occurrences that seem to conflict with one’s viewpoint. We have samples of both uses here:
 Marlow: What, my good friend, if you gave us a glass of punch in the meantime?
Hardcastle: Punch, Sir!
Marlow: Yes, sir, punch! A glass of warm punch, after our journey, will be comfortable. (Oliver Goldsmith, She Stoops to Conquer 1773: 24)
Hardcastle is taken aback at being ordered around in his own house; he repeats the requested item as if he had not heard right, and Marlow repeats it twice to reassure him and to re-affirm the request. In a like manner, recurrence can be used in REPUDIATION as defined by Halliday and Hasan (1976): rejecting some material stated (or implied) in the previous discourse. The material is repeated to show exactly what is being rejected, e.g. (Deeping 1930:729):
 “I think I told you that my name is Burnside.” “It might be Smith, sir, or Jones, or Robinson. “It is neither Smith, nor Jones, nor Robinson.”
In this dialogue, Mr Burnside is combating the other man’s attempts to brush aside his identity with trivial, everyday names. Still another contextual factor eliciting recurrence is the need to overcome irrelevant interruptions and get on with a statement:
 Hardcastle: He first summoned the garrison—
Marlow: Don’t you think the ventre d’or waistcoat will do with the plain brown?
Hardcastle: He first summoned the garrison, which might consist of about five thousand men—
Hastings: I think not; brown and yellow mix but very poorly.
Hardcastle: I say, gentlemen, as I was telling you, he summoned the garrison, which might consist of about five thousand men— (Goldsmith 1773: 23).
14. In poetic texts, the surface organization of the text is often motivated by special correspondences to the meaning and purpose of the whole communication.17 In Tennyson’s (1930: 237) well-known lines:
 Break, break, break
On thy cold grey stones, O Sea!
echoed in a later stanza with slight variation, the recurrences enact the motion of the waves being described. Frost (1969: 224) closes a poem with the lines:
 And miles to go before I sleep,
And miles to go before I sleep.
in order to evoke the even, continual motion of the speaker’s journey in a sleigh through a snowy landscape at night. Uses like the above are instances of iconicity: an outward resemblance between surface expressions and their content.
15. In all of our illustrations [20-26], the recurring expression kept the same reference, that is, it continued to designate the same entity in the “world” of the text (or discourse).18 Hence, stability was strongly upheld with obvious continuity (cf. III.14). If a recurrent expression has a quite different reference, the result can be disturbing, e.g. (Wilton Times, cited in Levin & Goldman 1978: 1):
 The bad news didn’t surprise Miss Ankrom, who is expecting a baby. She said she had been half expecting it.
Here, the lexical recurrence is not correlated with conceptual recurrence, the element ‘expect’ being used in two different senses (cf. V.1f.).19 The pronoun ‘it’ is non-determinate and might be carrying forward either ‘news’ or ‘baby’. Although the latter alternative is bizarre, it forces itself on the receiver’s attention because of the recurrence.
16. PARTIAL RECURRENCE entails using the same basic word components but shifting them to a different word class (compare the device of “Polyptoton” in classical rhetoric). In this fashion, an already activated concept can be re-used while its expression is adapted to various settings. Here are examples from the American Declaration of Independence:
[28.1] ... to assume among the powers of the earth the separate and equal station ... the causes which impel them to the separation.
[28.2] Governments are instituted among Men, deriving their just powers from the consent of the governed.
[28.3] mankind are more disposed to suffer, while evils are sufferable ... Such has been the patient sufferance of these Colonies
In his survey of partial recurrences like these, Dressler (1979a) notes that the presence of the one expression allows the other(s) to be rare or entirely novel. He cites the usage in Joachim Ringelnatz’s story about The ‘Whales and the Stranger:20
 The famous skyscraper made of toasted banana peels . . . south of the banana-scraper
The expression ‘banana-scraper’ would hardly be intelligible in the sense intended here without recourse to the co-referent expressions before it. Dressler also notes a story by Erich Fried (1975) where the title of Turtle-Turning and the expression ‘turtle-turner’ are introduced without explanation until a later passage:
 Everywhere he finds a helpless turtle fallen on its back, he turns it over.
17. Recurrence has the disadvantage, noted in IV.13, of reducing informativity. Therefore, techniques are often used in which forms recur with somewhat different content, or content recurs with different forms. PARALLELISM entails reusing surface formats but filling them with different expressions.21 In the Declaration of Independence, the British King is described as a walking disaster zone:
 He has plundered our seas, ravaged our coasts, burnt our towns.
Here, a series of similar though not identical actions, are expressed in parallel clauses (verb possessive pronoun direct object) with a recurrent ‘our’ in the middle of each. In another passage of the same document, the king’s various actions are all expressed via present participles preceded by ‘for’:
 For quartering large bodies of troops ... For protecting them ... For cutting off our trade ... For imposing taxes ... For depriving us ... For transporting us ... For abolishing the free System ...
Once again, there is some relatedness among these actions (all abuses of power) which is emphasized by the parallelism of form. In addition, the repetition of formats evokes the repetition of the king’s actions; indeed, the expression ‘repeated’ is itself repeated shortly after:
 Our repeated Petitions have been answered only by repeated injury.
By the same token, a reversal of form can stress a reversal of content, as in the closing words of the cited passage:
 We must hold them Enemies in War, in Peace Friends
18. PARAPHRASE is the recurrence of content with a change of expressions, as illustrated in this passage (Beerbohm 1955: 56ff.):
 I had never seen a murderer ... the decent symbol which indemnifies the taker of a life.
While  shows the paraphrase of a single concept (‘murderer’),  illustrates the paraphrasing of a more complex configuration (Govinda 1976: 206):
 When God became conscious of his omniscience, he suddenly felt terribly bored, because, whatever happened, he knew the outcome. There was no more any surprise; there was nothing that was not known beforehand.
It is not certain that the content of these underlined passages is the same. The question of paraphrase ultimately merges into the much debated question of synonymy.23 There seem to be only a few natural language expressions whose virtual meanings are identically the same. But there are many cases where contexts of occurrence determine the actual meanings (i.e., senses, cf. V.1) sufficiently that synonymy appears to be fulfilled, e.g. in  and 
19. Situationality can affect the outlook adopted on paraphrase and synonymy. Legal discourse, for instance, is intended to define certain kinds of behaviour beyond all doubt; accordingly, paraphrase is used richly in hopes of capturing every possible aspect of the intended content. The Gainesville Telephone Directory (1978-9: 16) quotes “Laws of Florida” which forbid using the telephone to make
 any comment, request, suggestion, or proposal which is obscene, lewd, lascivious, filthy, or indecent.
Under normal conditions, each series-’request/suggestion/ proposal’, and ‘obscene/lewd/lascivious/filthy/indecent’ would be taken as having elements meaning more or less the same (indeed, it might be hard to define one member of such a series without using other members in the definition). Still, the law is intended to cover all possible shadings and varieties of such behaviour, and thus prefers to be repetitious and pedantic if necessary. Shakespeare’s Constable Dogberry provides an immortal parody of this legalistic tendency:
 Marry, sir, they have committed false report; moreover, they have spoken untruths; secondarily, they are slanderers; sixth and lastly, they have belied a lady; thirdly, they have verified unjust things; and, to conclude, they are lying knaves. (Much Ado about Nothing V i 224-9)
The effect of this passage rests both on knowledge of the text type being spoofed (intertextuality supporting parody, cf. I.22) and on the failure of the Constable to maintain it (confusing the numbers, ending with a non-legal term ‘lying knaves’). A textual discrepancy is matched with a discrepancy normally accepted in the “real world” (cf. IX.8; X.16).
20. Our examples should suggest some kinds of motivations that call for recurrence, partial recurrence, parallelism, and paraphrase. In general, these techniques are deployed in order to insist upon relationships among elements or configurations of content within the text, most often equivalence (but opposition can be stressed also, as in sample .24 It follows that these techniques will be used above all in situations where stability and exactness of content can have important practical consequences, as in the application of legal texts to real life. Not surprisingly, text producers will strive to make texts fully determinate whenever a potential group of receivers is likely to contest detailed points. For instance, this passage is taken from a union contract:25
 Except for discoveries or inventions made during the course of approved outside employment, a discovery or invention which is made in the field in which the investigator is employed by the university or by using university funds, facilities, materials, equipment, personnel, or proprietary technological information, is the property of the university, and the inventor shall share in the proceeds therefrom.
We see a veritable battery of the devices we have discussed above: recurrence (‘discoveries/discovery’, ‘inventions/ invention’, ‘made/made’, ‘university/university/university’), partial recurrence (‘invention/inventor’, ‘employment/ employed’), and paraphrase (‘discovery/invention’, ‘investigator/inventor’, ‘facilities/materials/equipment’). Note that the text could have been clearer yet with parallelism:
[39a] a discovery or invention which is made in the field ... or which is made with the use...
21. Everyday communication does not demand this degree of certitude most of the time. More often, cohesive devices are used which shorten and simplify the surface text, even though, along the way, there is a certain loss of determinacy (cf. IV.29; IV-37). One obvious device is the use of pro-forms: economical, short words empty of their own particular content, which can stand in the surface text in place of more determinate, content-activating expressions (cf. Karttunen 1969; Padučeva 1970; Dressler 1972a: 27). These pro-forms free text users from having to restate everything in order to keep content current in active storage (cf. IV.2; V.4). The best-known pro-forms are the pronouns which function in the place of the nouns or noun phrases with which they co-refer (i.e. share reference in the sense of IV. 15).26 In this well-known children’s rhyme:
 There was an old woman who lived in a shoe.
She had so many children, she didn’t know what to do.
the pronoun ‘she’ makes it unnecessary to keep saying ‘the old woman who lived in a shoe’, ‘the old woman’, or even ‘the woman.
22. Sample  illustrates anaphora: Using a pro-form after the co-referring expression (cf. Postal 1969; Bresnan 1971; Edmonson 1976; Hankamer & Sag 1976; Kaplan 1976; Bullwinkle 1977; Camarazza et al. 1977; Webber 1978). Anaphora is the most common directionality for co-reference, since the identity of the conceptual content being kept current is made plain in advance.27 However, anaphora can still be troublesome if there is a lengthy stretch of text before the pro-form appears (cf. V.35ff.). By then, the original elements could have been displaced from active storage and other candidates may be mistakenly called.
23. The use of the pro-form before the co-referring expression is called cataphora (cf. Halliday & Hasan 1976). Processing would require the creation of a temporarily empty slot a position on a hold stack, perhaps, in the sense of IV.8, until the required content is supplied. Such a mechanism would work best if the distance between the pro-form and the co-referring expression is kept within limits, e.g. inside the boundaries of a single sentence:
 I don’t know if he’s serious, but my roommate wants to walk a tightrope over Niagara Falls.
Sample , taken from a student’s essay, is not the only kind of cataphora. A pro-form may look ahead to an entire event rather than an individual object, as in Halliday and Hasan’s (1976: 56) example:
 I would never have believed it. They’ve accepted the whole scheme.
Also, cataphora can be used to generate uncertainty and therefore to intensify receivers’ interest (cf. VII.13). One story starts off like this:
 He was scarcely ten years old when he was first arrested as a vagabond. He spoke thus to the judge: “I am called Jean François Leturc...” (François Coppée The Substitute, in Coppée 1891: 91).
A detailed self-description of the lad’s life follows, removing all doubt about the identity carefully withheld in the opening sentence. Receivers will be stimulated to find out how a ten-year-old of whatever identity comes to be arrested—problematic knowledge because apparently not easy to connect (cf. III.17).28 The cataphora raises a momentary problem in the surface text and helps to propel the readers into the story.
24. To test whether this interest effect can be empirically documented, an experiment with the ‘rocket’ text (to be depicted in more detail in IX.25ff) entailed, among other things, showing a group of readers this rearrangement of the original  in I.1:
[4c] Empty, it weighed five tons. For fuel it carried eight tons of alcohol and liquid oxygen. There it stood in a New Mexico desert: a great black and yellow V-2 rocket 46 feet long....
With the original opening sentence at the end of the paragraph, the pronoun ‘it’ in the new beginning is now cataphoric. The recall protocols did reveal a striking effect. Whereas only 30 per cent of the readers of the original recalled both fuels, 80 per cent did so after seeing the inversion. In exchange, 80 per cent reading the original recalled the colours, while only 30 per cent did so for the inversion. Apparently, the markedness of the inverted opening caused only a redistribution of the readers’ attention rather than an absolute increase. This finding corresponds to the mechanism known as the “von Restorff effect” concerned with special markedness (cf. Wallace 1965). Still, the usefulness of cataphora for creating focus on some block of content, in this case, impelling readers to use content heavily in trying to figure out the co-referent for ‘it’, does seem clear.29
25. Other elements besides nouns or noun phrases can be correlated with pro-forms. The verb ‘do’ is frequently employed as a pro-verb to keep current the content of a more determinate verb or verb phrase (cf. Karlsen 1959: 124ff ; Isačenko 1965: 172f.; Roggero 1968; Haskell 1973; Vater 1975: 37f; Halliday & Hasan 1976: 125ff.). In this instance (Goldsmith 1773: 36):
 miss hardcastle: I understand you perfectly, sir.
marlow (aside): Egad! and that’s more than I do myself.
the pro-verb ‘do’ stands for ‘understanding Marlow perfectly’. The pro-verb can, as we see, co-refer with a considerable block of content. In this text:
 To this day I am ashamed that I did not spring up and pinion him, then and there. Had I possessed one ounce of physical courage, I should have done so. (Beerbohm 1958: 57)
the term pro-modifier might be used for ‘so’, or, more specifically, pro-complement (cf. Steinitz 1968: 148ff ). The ‘so’ can stand for whatever modifiers were connected to the verb in the original verb phrase (cf. Bolinger 1970; Bouton 1970). In British English, ‘so’ is more often omitted in such usage than in American. In American English, the Inspector’s reply (Priestley 1980: 299):
 mrs birling: I don’t understand you, Inspector.
inspector: You mean you don’t choose to do, Mrs Birling.
would more likely be:
a inspector: You mean you don’t choose to do so, Mrs Birling.
Another pro-form would be a pro-modifier like ‘such’ (cf. Hasan 1968: 78; Palek 1968:6iff.; Figge 1971: 175), as in:
 Gerald Middleton was a man of mildly but persistently depressive temperament. Such men are not at their best at breakfast. (Wilson 1958: 3)
Here, ‘such’ co-refers with the entire depiction of the ‘man’.
26. It would be wrong to imply that pro-forms must always co-refer with elements of the same type, e.g. pronouns with nouns, pro-verbs with verbs, pro-complements with complements, and pro-modifiers with modifiers. Such correspondences are at best preferences which have the advantage of making already parsed grammatical frameworks re-usable. The pro-forms must also fit into the grammatical settings where they are needed. Consider this well-known passage (Julius Caesar I ii 194-5):
 Yond Cassius has a lean and hungry look.
He thinks too much; such men are dangerous.
The co-reference of ‘Cassius/he’ is entirely straightforward: noun followed by pronoun, both in the subject slot of their respective sentences. In contrast, the pro-modifier ‘such’ carries forward the content of ‘has a lean and hungry look’ plus ‘thinks too much’ two verb phrases thus co-referring with a single pro-adjective. Some researchers (e.g. Lakoff 1968) would also classify ‘men’ as a pseudo-pronoun or quasi-pronoun on the grounds that it has only minimal content of little relevance, criteria that would include ‘thing’ as well as equivalents in other languages (German ‘Ding’, Italian and Spanish, ‘cosa’, etc.) (cf. Green 1968: 25; Hasan 1968: 94f; Dougherty 1969: 513f.).
27. In addition, the pro-forms must often be correlated with entire clauses (“clausal substitution” in Halliday & Hasan 1976: 130-141). The pro-form ‘so’ is especially versatile. In this stretch of text:
 “Are you to get in at all?” said the Footman. “That’s the first question, you know.”
It was no doubt; only Alice did not like to be told so. (Carroll 1960: 51)
the ‘so’ carries forward the entire content of what the Footman said. In the following usage:
 “Of course you agree to have a battle?” Tweedledee said in a calmer tone. “I suppose so,” the other sulkily replied. (Carroll 1960: 241)
‘so’ signals affirmation of the other person’s utterance, and its converse would be ‘not’.
30. The “substitution of clauses” is carried out by pro-forms which signal that the content of the clauses is to be kept active, not their surface format.28 The settings of pro-forms also vary in regard to specificity. Lakoff (1968) suggests that the usual progress’ starts with the most specific and determinate content and moves toward the least. We might have a sequence of (a) proper name, (b) specific description, (c) general class (pseudo-pronoun in the sense of IV.26), and (d) pro-form. An illustration might be:
 Napoleon arrived at the palace. The conqueror of Austria was in high spirits. I never saw such an elated man. He hardly ever stopped talking.
This progression is probable because the content should be most clearly specified at the first use before re-use later on. However, a reversal of the progression could be a striking means to reveal the identity of the referent bit by bit. That tactic is in fact used in a story by Nikolai Leskov (196i: 55):
 Who should walk in but a venerable old man in whom His Grace immediately recognized one of the saints of the church, no other than the Right Reverend Sergius.
Sample  shows how effectiveness can be increased by not following a convention set up for the sake of efficiency (cf. I.23; III.9).
29. The efficiency criterion is stressed in Beaugrande (1980a) as a prime motivation for pro-forms in general (cf. IV.11). At a certain point, however, a trade-off appears between compactness and clarity. Pro-forms save processing effort by being shorter than the expressions they replace, but if those expressions are hard to locate or determine, the savings are lost again on search and matching operations. Various techniques can be applied in non-determinate cases. Chafe (1976: 47) suggests that a sample like:
 Ted saw Harry yesterday. He told him about the meeting.
would be processed with a preference for keeping the subject status constant (Ted = he, Harry = him).31 Another strategy would be to consult the organization of the situations, objects, or events in the textual world. When the Declaration of Independence says:
 He has constrained our fellow Citizens . . . to become executioners of their friends and Brethren, or to fall themselves by their Hands.
the pro-form ‘their’ is shifted in rapid succession from ‘Citizens’ to ‘friends and Brethren’; any other reading would not fit the events. The same factor applies to the ‘rocket’ text that begins:
 A great black and yellow rocket stood in a New Mexico desert. Empty it weighed five tons.
From the standpoint of syntax alone, ‘it’ might co-refer with ‘rocket’, ‘desert’, or even ‘New Mexico’. The lexicon would not help, since no reasonable definitions would stipulate what a rocket, a desert, or the state of New Mexico ought to weigh. The co-reference is simply resolved via the world-knowledge that the weight of a flying object such as a ‘rocket’ is problematic32 (it may cause a flight to fail, cf. III.17) and hence likely to be mentioned; parts of the landscape are seldom moved, so that their weight would be irrelevant for normal tasks (and probably undiscoverable anyway).
30. The preference for problematic knowledge in textual discourse is a pervasive principle of organization because it determines what people consider interesting and hence worth talking about (cf. Schank 1977; Beaugrande 1980a). Consider this fragment of a conversation:
 “Next morning he gets up, has a fire lit, orders in three shillins’ worth o’ crumpets, toasts ‘em all, eats ‘em all, and blows his brains out.” “What did he do that for?” inquired Mr Pickwick abruptly (Dickens 1836-37:617)
The pro-forms ‘do that’ might refer to all of the mentioned actions in Sam’s story (‘gets up’, ‘has a fire lit’, etc.), but there is little doubt that Mr. Pickwick’s question is only directed to the last-mentioned one. ‘Blowing one’s brains out’ is by far the most problematic, since no reason is apparent, whereas people ‘get up’, ‘light fires’, ‘toast crumpets’, etc., in the everyday course of life.
31. The same principle would apply to sorting out homonyms (words of the same outward format but differing in meaning or function), such as the pro-form ‘one’ and the numerical ‘one’. The pro-form is often used for an unspecified member of a class, e.g.:
 It’s a very distressing case very; I never knew one more so. (Dickens 1836-37: 128)
where ‘one’ designates any (unknown) ‘case’ meeting the description ‘more distressing than the present case’. Similarly, ‘one’ can serve in place of an unspecified person, e.g. (Govinda 1976: 15):
 One should not form judgements on the ground of such perceptions, nor should one allow one’s thoughts to be determined and led by them.
Now consider this newspaper headline (Gainesville Sun 20 Dec. 1978):
 San Juan Gunfire Kills One
Text receivers will hardly construe ‘one’ as an unspecified person (‘San Juan Gunfire Kills People’), since that would not be informative: it entails no problem because gunfire can be fatal in any city. Hence, ‘one’ will be preferentially taken as the numerical, being the number of entities fatally affected by this particular shooting incident.
32. Another cohesive device contributing to compactness and efficiency is ellipsis (cf. Karlsen 1959; Gunter 1963; Isačenko 1965; Crymes 1968; Dressler 1970; Halliday & Hasan 1976; Grosz 1977). An examination of the sources just cited will reveal considerable dispute over what constitutes ellipsis, due to differences in the requirements of a grammar. If the criteria for well-formedness and logical stringency are very extensive, a great many real texts will appear elliptical.33 In the procedural approach advocated here, ellipsis is present only when text processing involves an apperceptible discontinuity in the surface text. The question of whether a given sample is truly elliptical must eventually be decided empirically (which surface structures do text users consider discontinuous?).
33. Usually, ellipsis functions via a sharing of structural components among clauses of the surface text. The typical case is anaphoric, i.e. the complete structure occurs before the elliptical one (cf. IV.22):
 Mine be thy love, and thy love’s use their treasure (Shakespeare Sonnet XX)
In , the verb ‘be’ from the first clause is needed to complete the second (‘thy love’s use be their treasure’). The complete structure should still be recoverable in such cases, so that the distance to the elliptical one must be kept within limits. Ellipsis does, however, often occur in a new utterance unit rather than in the same one:
 The daughter is said to be well-bred and beautiful; the son an awkward booby, reared up and spoiled at his mother’s apron strings. (Goldsmith 1773: 14)
A change of speaker may be involved:
 Brutus: Let Me tell you, Cassius, you yourself
Are much condemned to have an itching palm;
To sell and mart your offices for gold
Cassius: I an itching palm? (Julius Caesar IV iii 9-12)
The recovery of the full forms (‘the son is said to be an awkward booby’, ‘I am much condemned to have an itching palm?’) is not difficult, even though, in , there is some distance between the complete stretch and the elliptical one.
34. Ellipsis is most noticeable when follow-up structures lack the verb a relationship called gapping by Ross (1970)—because in English at any rate, the verb is the least dispensable element in a clause. The ellipsis of subjects in independent clauses is not uncommon:
 He’s always asleep. Goes on errands fast asleep ... I’m proud of that boy—wouldn’t part with him on any account. (Dickens 1836-37: 55)
The dispensability of subjects may be related to Chafe’s observation about subject roles (cf. IV.29); the subject slot is the likely place to direct attention when completing elliptic structures like ‘goes on errands’ because this role is assumed to remain stable. All the same, Leech and Svartvik (1975: 168) note that ellipsis of subjects in dependent clauses is rare in English, e.g.
 He was so tired that went to sleep.33a
even though there would be no trouble supplying the subject.
35. The ellipsis of subjects or other dispensable elements illustrates the complexity of interaction between cognition and syntactic conventions. The identity of the missing subject is beyond all doubt in ; yet such a construction is seldom used. A procedural approach is concerned with discovering the conditions under which ellipsis becomes frequent and why heavily elliptic texts are nonetheless comprehensible. The texts of Mr Alfred jingle, though admittedly eccentric in their fragmented format, are not hard to follow:
 Fired a musket—fired with an idea—rushed into wine shop—wrote it down—back again—whiz, bang—another idea—wine shop again—pen and ink—back again—cut and slash—noble time, sir. (Dickens 1836-37: 11)
Yet if the function of syntax in communication is, as we argued in IV.2ff, to provide a surface organization that constrains hypotheses about the organization of underlying concepts and relations (a system with few options acting as a kind of “distant early warning” for a system with many more options), then partial use of syntax, as in , would constitute a substantial processing strain. The problem-solving that imposes cohesion and coherence on discourse (see Chapter III) would have to work unusually hard in all kinds of directions. Mr Jingle’s utterances are easy enough to piece together when supplied in print, but they could be confusing if we heard them spoken.
36. Little research has been done on the processing of elliptical texts, because the well-formed sentence was usually taken as the obligatory unit for language experiments.34 The dominant role of the sentence in linguistic theories engenders the notion that “perhaps all utterances are derived from implicit complete sentences” (R. Brown 1973: 209). However, this notion is hardly convincing and certainly not empirically proven. The conversion of a text such as  to complete sentences could be useful, but is not necessary for processing. Moreover, it would be difficult to agree upon any complete version. Quite plausibly, even ordinary processing might bypass some utilization of syntax whenever the expenditure of effort would outweigh the benefits; the processor would attempt to recover coherence more directly, doing only “fuzzy parsing” of the surface (cf. Burton 1976; VII.9). Exhaustive utilization of syntax, possibly done by augmenting the transitions of the grammar net (cf. IV.9), would be encouraged whenever other cues do not prove conclusive, e.g. in the presence of ambiguities.
37. Like the use of pro-forms, ellipsis illustrates the trade-off between compactness and clarity (cf. IV.29). Utilizing texts with no ellipsis consumes time and energy. At the other extreme, very heavy ellipsis cancels out any savings of time and energy by demanding intensive search and problem solving. Text users must weigh the appropriateness of ellipsis to the setting to decide what extent will contribute to rather than damage efficiency (cf. III.9). This weighing operation is a typical difference between an abstract system of syntax and a procedural model of syntax in interaction with other factors of textuality.
38. Cohesion is further supported by tense and aspect (cf. Reichenbach 1947; Weinrich 1964; Wunderlich 1971; Dowty 1972).35 These categories are organised very differently in various languages (cf. Dressler 1972a: 47ff ) Usually, there are means to distinguish: (a) past, present, and future times; (b) continuity vs. single points; (c) antecedent vs. subsequent; (d) finished vs. unfinished. Some of these distinctions arise mainly from the perspective of the text users at that moment (e.g. past, present, and future are relative to the situation), and others from the organization of text-world situations or events among themselves. When the verb systems do not make the distinctions explicit, modifiers or junctives should be used.
39. The strategies of text formation reflect some influences of the order in which tenses and aspects are used. In Hebrew, there is a sequence of tenses which must be used consecutively (Harweg 1968: 284). In Bahinemo, a language of Papua New Guinea, the verb of a single dependent clause at the beginning sets the time for all events and situations mentioned in the paragraph (Longacre 1970). In Godié of the Ivory Coast, the time need be set only once for an entire text (Grimes 1975: 23 2). In Xavante of Brazil, two distinct aspect systems are used for events vs. non-events (Grimes 1975: 93). In Mumuye and Longuda of Nigeria, a progressive aspect is used for settings as opposed to the main sequence of events (Grimes 1975: 234).
40. Such striking variety points up the enormous complexity and subjectivity involved in the organization of time in a textual world (cf. Bruce 1972). The view that time is passing at a steady rate (a pre-Einsteinian, but commonly held view) is far less decisive than the interrelationships among situations and events that differ not only in duration, but also in structure and importance. As Talmy (1978: 21) points out, the same event can be expressed in different perspectives, e.g.:
[65a] The beacon flashed.
[65b] The beacon kept flashing.
c The beacon flashed five times in a row.
In [65a], the event is seen as a closed unit at a single point in time. In [65b], the event is a multi-part unit extending over an unbounded expanse of time. In [65c], the event is a multi-part unit with defined time boundaries. Fillmore (1977: 74) suggests that “any particular verb or other predicating word assumes, in each use, a given perspective” on a “scene”. The need for envisioning “scenes” in order to process even syntactic surface formats (e.g. anaphora) has been eloquently illustrated by Dillon (1978: 70ff.).
41. Still, some consistent principles of time organization might be discoverable and relatable to systems of tense and aspect. If textuality rests on continuity, as claimed in III.14, text users would naturally see text-world events and situations as related. Noticeable gaps could be filled by updating, that is, making inferences (in the sense of I.11; cf. V.32-34) about how the text-world is evolving (cf. Sacerdoti 1977: 15; Winston 1977: 386). For instance, when the rocket ‘rises’ in sample , we update its location from ‘desert’ to ‘sky’, its fuel supply from ‘eight tons’ to less, etc., without needing explicit statements. Some further principles of time organization have been proposed by Leonard Talmy (1978): (a) plexity, the capacity of having multiple parts; (b) boundedness, the capacity of having discernible limits; (c) dividedness, the lack of internal continuity; and (d) distribution, the pattern of actions/events in a time unit. Though considered by both Halliday and Talmy to be “grammatical” notions, they are clearly traceable back to human cognition about events and situations (cf. Miller & Johnson-Laird 1976). As in many other domains, the cohesion of the surface text rests on the presupposed coherence of the textual world (cf. Morgan 1978).
42. A clear device for signalling the relationships among events or situations is junction, the use of junctive expressions (in traditional grammars rather indiscriminately all called “conjunctions”) (cf. Gleitman 1965; Dik 1968; Tai 1969; Harweg 1970; Dougherty 1970-71; R. Lakoff 1971; Halliday & Hasan 1976; Lang 1976; van Dijk 1977b). At least four major types should be discussed:
(a) Conjunction links things which have the same status, e.g., both true in the textual world.
(b) Disjunction links things which have alternative status, e.g., two things of which only one can be true in the textual world.
(c) Contrajunction links things having the same status but appearing incongruous or incompatible in the textual world, e.g., a cause and an unanticipated effect.
(d) Subordination links things when the status of one depends on that of the other, e.g., things true under certain conditions or for certain motives (precondition/event, cause/ effect, etc.).
43. These types are recognizable by the classes of junctives as surface cues for each. Conjunction is most often signalled by ‘and’, and less often by ‘moreover’, ‘also’, ‘in addition’, ‘besides’, ‘furthermore’, etc. Conjunction is an additive relation, for instance, when connecting two interdependent events or situations mentioned within a sentence, e.g.:
 The great birds like to roost in trees in parks just outside the town, and since 1885 the local citizens have made the best of the situation. (TIME 26 March 1979)
Conjunction can carry across the boundaries of the sentence:
 Sadat called this a means of protecting the “human rights” of the Gaza Palestinians. And to ensure that Gaza attains autonomy, Sadat wanted a firm commitment. (TIME 26 March 1979)
Conjunction can link utterances not formatted as complete sentences at all, provided an additive or interdependent relationship obtains:
 After all I’ve done for law enforcement and for them to treat me this way. (TIME 26 March 1979)
Conjunction is the default junction, since, unless specified otherwise, events and situations are combined additively in a textual world. There is no motive to place ‘and’, ‘also’, ‘in addition’, etc. between all clauses or sentences; in fact, such a practice renders the text dull except for occasional special effects (compare the device of “polysyndeton” in classical rhetoric). The use of such junctives is more likely when interdependency is not obvious and should be stressed.
44. Disjunction is nearly always signalled by ‘or’ (sometimes expanded to ‘either-or’, ‘whether or not’, etc.). It is most commonly employed within sentences:
 A man must not be too precipitate, or he runs over it [his hat]; he must not rush to the opposite extreme, or he loses it altogether. (Dickens 1836-37: 49f.)
Within a sentence, ‘or’ joins alternatives both of which are current in active storage, but only one of which obtains in the textual world. Between sentences, ‘or’ tends rather to announce an afterthought, an alternative not considered before:
 “Unless Mr Winkle feels aggrieved by the challenge; in which case, I submit, he has a right to satisfaction.” Mr Winkle, with great self-denial, expressed himself quite satisfied already. “Or possibly, “ said the man, “the gentleman’s second may feel himself affronted.” (Dickens 1836-37: 31)
Disjunction may not be so easy to process, since text users would have to carry forward both alternatives in active storage until a resolution is found.
45. contrajunction is signalled most often by ‘but’ and less often by ‘however’, ‘yet’, ‘nevertheless’, etc. It is the function of contrajunction to cause problematic transitions at points where seemingly improbable combinations of events or situations arise. In a sample like:
 Discouraged aides talked openly of the trip becoming a debacle. But at the last minute Carter achieved a victory of presidential diplomacy. (TIME 26 March 1979)
the text producer deploys ‘but’ to alert receivers that the expected ‘debacle’ became something totally different, a ‘victory’. In this text:
 Carter was upset and angry. But Begin remained firm. (TIME 26 March 1979)
contrajunction signals that a natural response to the anger of a powerful personage, namely conciliation, was not the case.
46. Subordination is represented by a large repertory of junctive expressions: ‘because’, ‘since’, ‘as’, ‘thus’, ‘while’, ‘therefore’, etc. Subordinating junctives make common types of coherence relations explicit, such as those outlined in I.6-11 (cf. also Chapter V). One type well represented by junctives is cause (necessary conditions, cf. 1.7):
 It would befoul Long Beach Harbour with oil spills and seriously worsen the local smog problem, because merely unloading the oil would release hydrocarbon fumes into the atmosphere. (TIME 26 March 1979)
Reason (rational human reaction, cf. 1. 8) is also frequent:
 The judge refused, on the grounds that he lacked authority.
47. The repertory of ‘junctive expressions is large for the i relation of temporal proximity (cf. I.10) as well: ‘then’, ‘next’, ‘before’, ‘after’, ‘since’, ‘whenever’, ‘while’, ‘during’, and so on. Proximity can be sequential if events or situations are ordered in respect to each other, e.g.:
 The President emotionally declared that he was “glad to be home”. Then he told the gathering what it had come to hear. (TIME 26 March 1979)
Overlap can be indicated rather than sequentiality:
 The following day the Egyptian Cabinet also unanimously approved the final details of the agreement. Meanwhile, the Israeli and Egyptian Defense Ministers met in Washington. (TIME 26 March 1979)
Temporal proximity can involve a chaining where the termination of one event or situation coincides with the initiation of the next, possibly with causality implied as well.
 When Carter brought up Sadat’s proposals, Begin said they were “completely unacceptable”. (TIME 26 March 1979)
48. Still another use of subordination is to signal modality, that is, the probability, possibility, or necessity (or the opposites of those) of events and situations (cf. Reichenbach 1976). The junctive ‘if’ marks a condition under which some event or situation would be true, e.g. in Mrs Thatcher’s campaign remark (Daily Telegraph 26 April 1979):
 We can have German standards of living if we have German standards of work.
Modality is important for projected events and situations, those which might happen or might have happened in a textual world (cf. V. 28). For past time, there is usually no longer any possibility for the projection to be true:
 If the principle tower of Rochester Castle had suddenly walked from its foundation, and stationed itself opposite the coffee-room window, Mr Winkle’s surprise would have been nothing compared with the profound astonishment with which he heard this address. (Dickens 1836-37: 24)
The contrary-to-fact status of the walking tower and its results are, as we see, also signalled by marked verb tenses (‘had walked’, ‘would have been’).
49. The intricacies of junction are far greater than our sketch might imply. Except for disjunction, the use of junctives as explicit signals is rarely obligatory, because text users can recover relations such as additivity, incongruity, causality, etc. by applying world-knowledge. We could delete the junctives from samples , , , , and , adding punctuation occasionally, without rendering the texts doubtful. But by using junctives, text producers can exert control over how relations are recovered and set up by receivers. For instance, using ‘then’ in  makes it clear that the President’s ‘emotional declaration’ was not (as might be assumed if ‘then’ were deleted) what ‘the gathering had come to hear’; the producer can thus insert his or her own interpretation into the monitoring of the situation (cf. VIII.1).
50. In this perspective, junction demonstrates how communicative interaction, not just grammatically obligatory rules, decides what syntactic formats participants use. Junctives can be a simple token of courtesy to help make reception of a text efficient. They can assist the text producer as well during the organization and presentation of a textual world. They can, as we saw in IV.49, imply or impose a particular interpretation. Yet they are seldom to be found in every transition among events and situations of an entire textual world. Apparently, a certain degree of informativity is upheld by not using junctives incessantly. There are other surface categories which can fulfil the same functions, e.g. using causative verb forms (cf. Grimes 1964 for a comparison of Huichol and English) or inserting interjections (cf. Gillich 1970; Franck 1979).
51. A special aspect of interaction between syntax, informativity, and communicative settings has been stressed in functional sentence perspective mentioned in II.18. The mere placement of materials in the earlier or later stretches of clauses and sentences suggests the relative priorities and degrees of informativity of underlying content (for discussions and surveys, cf. Mathesius 1928; Firbas 1962, 1964, 1966, 1968, 1974, 1975; Halliday 1967-8; Beneš 1968; Chafe 1970, 1976; Sgall et al. 1973; Daneš (ed.) 1974; Dahl (ed.) 1974; Grossman, San & Vance (eds.) 1975; Grimes 1975; Firbas & Golková 1976; Li (ed.) 1976; Jones 1977). The extent to which this aspect controls syntax varies according to the number of other constraints that apply. In English, the lack of a differentiated morphemic system in many areas places heavy constraints on word-order patterns. In Czech, with its richer morphemic systems, word order can follow the functional perspective much more faithfully (Sgall et al. 1973).36
52. Since people tend to give a point of orientation before presenting new or surprising things, informativity tends to rise toward the end of a clause or sentence. Consider the opening of a story from the Tibetan biographies of the Eighty-Four Siddhas (Govinda I 1976: 2 5).
[80.1] There once was a hunter called Savari. [80.2] He was very proud of his strength and marksmanship. [80.3] The killing of animals was his sole occupation, [80.4] and this made his life one single sin. [80.5] One day, while he was out hunting, he saw a stranger.
The text begins with an empty expression ‘there once was’ that merely asserts the existence of the main character. The character’s profession and name are saved until the end stretch, with the more specific name following the less specific profession. The character now can appear as the subject of the next sentence, whose predicate offers more material about his traits. Of course, a hunter could be readily expected to possess ‘strength’ and ‘marksmanship’; but in this story, the hunter will be defeated on precisely those two counts, such that the narrator is motivated to stress them here. A format such as:
[80.2a] His strength and marksmanship made him very proud.
would create less focus on the crucial talents In [80.3], ‘the killing of animals’ makes a good beginning because it follows determinately from the already stated profession ‘hunter’; the new material is that this activity is his ‘sole occupation’ (i.e. unredeemed by works of kindness), leading into [80.4]. An anaphoric ‘this’ (in [80.41) keeps the content of the preceding sentence active so that a characterization (‘a single sin’) can be added. The opening of [80.5] presents Savari in his usual occupation; the sentence ends up with the expression of a new arrival (‘stranger’) on the scene.
53. We shall look into informativity in more detail in Chapter VII. We wanted to note here that, since cohesion rests on the assumption of underlying coherence (cf. IV.41), the sequencing of surface texts gives signals about the shared knowledge to be applied during a given stage of the communicative interaction. For example, due to the strategic usefulness of presenting known material first, the subjects of English sentences are often, though certainly not always, expressions (re)activating established or predictable content (cf. Firbas 1966a). The latter stretch of the predicate is, in turn, especially serviceable for creating focus.
54. A subsidiary cohesive system available only for spoken texts is intonation (cf. Halliday 1967; Crystal 1969; Lehiste 1970, 1975; Brazil 1975). In English, the usual pattern is a rising intonation toward the ends of clauses or sentences, notably reaching a peak on the last expression conveying substantive content. Although research was long centred on clauses and sentences, David Brazil (1975) has recently undertaken an account of intonation in whole texts or in texts within discourses. He adopts Halliday’s (1967) “tones” and relates them to types of discourse actions (cf. VI.11). Invoking (called “referring”) is done when the speaker presents predominantly known or expected material, while informing (called “proclaiming”) is done when the speaker presents predominantly new, unexpected, corrective, or contrastive material (cf. VIII.10). Hence, informing is more prone to elicit responses from other participants than is invoking. There is also a neutral option not deemed to qualify as either action.
55. The tone is the rising or falling tendency of a tone group (a stretch of text uttered as a unit). The basic choice is between a falling tone and a falling-rising (failing, then rising) tone (Tones 1 and 4 in Halliday’s scheme). The falling tone is normally used for informing, and the falling-rising for invoking. If we use arrows pointing either down (fall) or down, then up (fall-rise), we could have four patterns for the same two-part utterance, as in Figure 5 (examples from Brazil 1975: 6)
[81a] would be used if the hearer is assumed to know about the speaker’s reading ‘Middlemarch’, but not about future plans. [81b] would be appropriate if the reading of ‘Adam Bede’ were already known, but not the intended occasion or time of the reading. The reversing of the clauses, intriguingly enough, does not alter this outlook: [81c] would work like [81a], and [81d] for [81b]. Apparently, the contrasting intonation of the clauses makes the rising-falling one appear as background material and the falling one as foreground, irrespective of their order of utterance.
56. In addition, Brazil (1975: 7f.) identifies two marked or intensified options indicating an extra measure of involvement on the part of the speaker. An intensified informing action would have a rising, then falling tone (Halliday’s Tone 5). If placed on the opening clause of [81b], rise-fall tone would stress the time of completing ‘Middlemarch’ (i.e. then and only then, then and not a moment sooner, etc.). Similarly, an intensified invoking action would have a simple rising tone (Halliday’s Tone 2). If applied to the second clause of [81b], the effect would be to turn the utterance into a question, or into a statement seeking support or confirmation in view of the speaker’s own uncertainty. The rising tone is especially suited for an insistent question, or, in ‘if’-clauses (cf. IV.48), an insistent condition. Finally, Brazil recognizes a low rising tone (i.e. rising only from low to mid key) (cf. IV. 57) as a “neutral” option used to avoid committing oneself to one type of discourse action (Halliday’s Tone 3).
57. This basic scheme is combined with a differentiation of keys going all the way back to Henry Sweet (1906). Mid key is the pitch considered normal for the circumstance, and high key and low key the pitches above and below the norm, respectively. Brazil argues that the usual discourse sequence is high-mid-low, because the high key suggests an intention to continue the current stretch of discourse, and the low key an intention to end. In particular, high key is prominently employed in contrasts, either between two stated chunks of material or else between the stated material and what might be expected. Conversely, low key suggests equivalence of a chunk with a previous or expected chunk; stability is signalled by articulation with minimal effort. In an exchange like (Brazil 1975: 28):
[82.1] Where is he now? [82.2] In bed.
a high key answer would suggest that the location were bizarre or scandalous, while low key suggests that it is only to be expected. Hence, the high key would encourage further discussion, whereas the low key would indicate that no more need be said. The mid key is neutral and noncommittal in this regard, and would therefore be used when one wishes to leave the option of continuing completely open.
58. Even our brief outline of Brazil’s scheme should reveal its important implications for the study of texts as human activities. Intonation not only links together spoken surface texts; it also serves to qualify the linkage of concepts and relations both within the textual world and between the textual world and shared prior knowledge. Consider just the simple mechanisms of recurrence and paraphrase noted in IV.12-19 and IV.18-19, respectively. When one participant repeats or paraphrases a text just presented by another, the further development of the discourse depends crucially on intonation. A high-key recurrence or paraphrase elicits further explanation or justification, such as Hardcastle’s exclamation ‘Punch!’ in sample . A low-key recurrence or paraphrase would merely signal that the prior text has been heard and understood. A mid-key recurrence or paraphrase leaves it up to the producer of the prior text to decide whether anything more should be added. In this fashion, the choice of key is a gauge of intentionality and acceptability as speaker and hearer attitudes about cohesion, coherence, and informativity. As such, intonation contours have noteworthy influence upon situationality (what is going on in a particular communicative setting) and intertextuality (how to frame your text in regard to other people’s texts in the same discourse). Moreover, any disregard for the demands of efficiency, effectiveness, and appropriateness can be immediately regulated; a high-key paraphrase would be used to respond to a severe disregard, while a low-key paraphrase would be used for a mild disregard.
59. This chapter has been devoted to the factors in text cohesion. We suggested that short-range stretches of surface structure are set up as closely-knit patterns of grammatical dependencies; long-range stretches, in contrast, could be handled by re-utilizing previous elements or patterns, economizing where possible. We progressed from cases where the surface occurrences simply happen again toward cases where greater compactness is attained. Recurrence entails the exact return of materials (IV.12-17). Partial recurrence involves different uses of the same basic language items (word-stems) (IV.16). Parallelism is found when structures are re-used with different materials in them (IV.17). Paraphrase obtains via approximate conceptual equivalence among outwardly different materials (IV.18-19). We argued that these four devices are preferentially deployed when text producers wish to preclude uncertainty or contest. For everyday use, other devices serve to compact the surface text: pro-forms are brief, empty elements that are employed to keep the content of fuller elements current and, where expedient, to re-use basic syntactic structures (or compacted versions of these) (IV.21-3I); Ellipsis allows the omission of some structural components, provided that a complete version is recoverable (IV. 3 2 7). Pro-forms and ellipsis evince a trade-off where compactness might become so extreme that no savings in effort are attained after all, because energy is drained away reconstructing things (IV.29ff.; IV.37).
60. We then progressed toward devices which overtly signal relations within or among events and situations of the textual world. Tense and aspect can signal relative times, boundedness, unity, order, and modality of events and situations (cf. IV.38-41). Junction offers explicit markers for relationships of additivity, alternativity, incompatibility, and subordination through causality, time, modality, and so on (IV.4.2-50). We concluded by reviewing the contributions of functional sentence perspective as a correlation between priorities of knowledge or informativity and the arrangement of words in clauses and sentences (IV.51-53); and of intonation as the imposition of characteristic audible contours of tone and key upon texts in discourse, providing major cues about expectations, attitudes, intentions, and reactions (IV.54-8).
61. Though by no means complete or exhaustive, our survey should make it clear why the notion of “text cohesion” is substantially broader than usual notions of “text syntax” or “text grammar.” The broadening arises from two factors: the operationalisation of syntactic or grammatical structures as configurations utilized in real time; and the interaction of syntax or grammar with other factors of textuality. Syntactic theories of the 1980s and 1960s were not intended to account for those two factors, so that we may be compelled to develop new theories rather than ‘just “revising” or “extending” standard ones. We hope at least that we have raised some issues of the
kind that new theories of text cohesion ought to encompass; and that we have set forth motives for building such theories in the wider context of human interaction.
1 As pointed out by David Johnson (1977: 153), standard sentence grammars, centred around notions like “dominance” and “precedence”, have made very little of relational linkage and dependency. “Relational grammar” (cf. Cole & Sadock (eds.) 1977) is intended to compensate for this neglect.
2 On pattern matching, see note 17 to Chapter III.
3 There are of course numerous other definitions of the sentence (survey in O’Connell 1977), many of them inconsistent and confused.
4 The “state” of a system is the point where operations are centred at a given moment. We can have “macro-states” or “micro-states” by adjusting our scope of operations toward larger or smaller. We can also have different types of states: grammar states in cohesion (cf. IV.6), knowledge states in coherence (V.31), plan states in intentionality (cf. VI.13f.), information states in informativity, states of objects and participants in situationality (see note 6 to Chapter VII), and so on. Like many theoretical notions, “state” also figures as an entity in textual worlds (cf. V.26(a)).
5 “Macro-states” could presumably join to yield a “macro-structure” in the sense of van Dijk (1979b); cf. II-37.
6 Note that the actual use may be different for the same element in different dependencies. For instance, ‘in a desert’ is a “modifier” of the “head” ‘stood’, but ‘desert’ is also a “head” for the “determiner” ‘a’ (see Fig. 4).
7 In principle, the transition network should be equally applicable to production and reception of texts, as has been logically and mathematically demonstrated by Simmons and Chester (1979). There would be some obvious distinctions in the type of search, since the producer is making the original decisions and the receiver only recovering them; but in both cases, linkages must be found and tested before use. We approach the matter here mostly from the standpoint of reception, since that is the use for which the formalism was designed in the first place (see references in IV. 5).
8 It may be better to subdivide modifier types, e.g. “adjective”, “adverb”, and so on; this question will have to be resolved empirically: do language users always make those distinctions? And if they do, what about cases that normal speakers might find hard to decide, e.g. ‘bright’ in ‘The moon shone bright’?
9 Cf. note 15 to Chapter 3.
10 As noted in IV-43, conjunction with ‘and’ is essentially additive: it usually joins elements of the same type or status.
11 For example, there may be several transitions being tested and traversed in parallel.
12 Sec note 7 to this chapter.
13 These alterations are called miscues (cf. Goodman & Burke 1973).
14 We do not suppose that people would build transition networks for the granunatical dependencies of whole texts (they would rather build up conceptual-relational networks after each stretch of text was processed, cf. Chapter V). But they might well have traces left of a previously built network that would make re-use easier than building something new.
15 On recurrences of other kinds, see the demonstration in VII.29-42.
16 Of course, limited repertories of language systems, especially phonemes, make a certain amount of recurrence unavoidable (cf. Werth 1976; Beaugrande 1978b). Such recurrences would hardly be noticed. The use of recurrence for insistence is illustrated in VI.18; VIII.24, 26.
17 As noted in IX.9, poetic texts are definable according to their reorganisation of mapping strategies onto the surface text. Receivers therefore focus more attention on recurrences of all kinds.
18 The discussions over reference among philosophers have been lengthy and seldom conclusive (survey in Lyons 1977: 174-229). The tendency is to try to explain all kinds of reference on the basis of the few marginal cases we can nail down, e.g. naming objects present. See V.40 for a different outlook, where reference is a factor of textual worlds, not of words.
19 The “sense” is the actual knowledge conveyed by a text element within its continuity of coherence (cf. V.1f.). Doubtful reference of course follows from doubtful sense but not necessarily vice versa (see note 18 to this chapter).
20 Where it can be done without distorting the demonstration, we provide our own translations of non-English samples and citations. The creative use of partial recurrence is in conformity with the “general theory” of creativity outlined in Beaugrande (1979c).
21 For more illustrations, cf. VII-34.
22 For more illustrations, cf. VII-37, 41, and VIII.24.
23 See for instance Hirsch (1975). Some illustrations are noted in VII.37.24 The notion of “equivalence” was central in descriptive linguistics (cf. II.2if.). In our usage, however, there is usually some degree of approximation; the important factor is the stability of an actual system when occurrences (or blocks of occurrences) belong to the same general type. Such is the case here.
25 From the Agreement between the Board of Regents, State University System of Florida, and United Faculty of Florida, 1978-1981 (no date or place of publication), p. 22.
26 We use the term “co-refer” because it is well established, but we retain the reservations expressed in note 18. Perhaps “co-sense” would be more fitting than “co-reference” in some instances.
27 On different usage in Samoan, compare Chapin (1970).
28 The tendency to prefer problematic knowledge as discourse material is stressed again in IV. 29f.; IX.14, 26.
29 There may also be effects of primacy (the first part of a presentation being favoured) at work (cf. Meyer 1977: 308f.). See IX.37. 5 and note 23 to Chapter IX.
30 On the many uses of ‘so’, cf. Halliday & Hasan (1976: 140) and accompanying discussion.
31 Reichman (1978: 290) points out an interesting case where a proper name is used rather than a pro-form, even though no confusion would be possible, apparently because the person mentioned is outside the focus of attention; the person in focus is referred to via a pro-form.
32 See note 28 to this chapter.
33 The question is: “elliptical in comparison to what?” (Coseriu 1955-56). Alfred Whitehead was prone to condemning natural language because of its incompleteness.
33a This construction is in fact used in Spenser’s Faerie Queene (I i 42): ‘but he againe shooke him so hard, that forced him to speake’. Dillon (1978: 118) classes such usage as “ungrammatical”.
34 The “cloze” procedure of omitting words periodically for test persons to supply is not really comparable to using ellipsis in spontaneous discourse.
35 In common usage, “tense” designates the inflection of verbs to show time relative to the act of discourse, while “aspect” subsumes the boundaries (beginning, completed) and duration of events as marked by verb inflections.
36 On many other languages, cf. Grimes (1975); Grossman, San & Vance (eds.) (1975); Li (ed.) (1976); Grimes (ed.) (1978).
Return to main page
to Text Linguistics