III  

 

The Procedural Approach

  

 

1. DESIGNING A MODEL

 

 

   1.1 Text production has been investigated relatively seldom (Fodor, Bever, & Garrett, 1974: 434; Goldman, 1975: 289; Osgood & Bock, 1977: 89; Rosenberg [Ed.], 1977a: xi; Levin & Goldman, 1978: 14; Kintsch, 1982: msp. 3).1 [“Msp.” indicates page numbers cited from a manuscript then in press.] It is not easy to control or to sort into component processes. As a task, text production is typically OPEN (adapting to a steadily evolving situation) rather than CLOSED (fixed for all situations) (cf. Poulton, 1957; Gentile, 1972; Schmidt, 1975; I.1.3); and OPEN-ENDED (not terminating definitively at a particular moment) rather than CLOSED-ENDED (with an obvious end) (Horowitz & Berkowitz, 1967: 207; cf. III.1.26; III.3.2.12). Text production is normally SPONTANEOUS (reacting to ongoing context) and EXTEMPORANEOUS (improvised, not previously prepared). When the production of one text follows upon reception of another, the two processing acts may be hard to sort out (Rosenberg, 1977b: 89). For all these reasons, it is unclear how far EXPERIMENTAL conditions can be comparable to the NATURAL conditions of text production (cf. II.2.8, 33; II.3.27; III.2.40; IV.1.9; IV.2.16, 21; V.1. 13f, 19; VI.1.29f; VI.3.20ff).

   1.2 In the experimental literature, most investigators presented test subjects a “stimulus” whose “response” was, or included, producing a text. The most common “dependent variables” whose quantity the experimenters were trying to predict were the test subjects’ (a) latency, also called reaction time (how long they take to get started); (b) time on task (how long they take to get finished); and (c) error count (how often they go wrong).2 [Errors were usually defined as failures to follow instructions, or as observable “speech disruptions,” e.g. false starts or ‘uh’s (cf. Mahl, 1956; Maclay & Osgood, 1959; V.3.12).] Some common experimental designs were: {88}

    1.2.1 Duplicating a message. Subjects have to memorize and then utter a whole sentence (Lackner & Levine, 1975). A variant with a shorter span is shadowing, where people immediately repeat an acoustically presented message (Cherry, 1953; Broadbent, 1958; Neisser, 1967). Early research indicated that shadowing worked from a direct auditory trace of the message, but soon, the impact of conceptual content (Treisman, 1969) and personal involvement (Moray, 1959) became evident.

    1.2.2 Using components in a message. The experimenter provides only key words that test subjects are to use in a sentence or utterance (Taytor, 1969; Gosnave, 1977). The sentence may have to define the key words (Reynolds & Paivio, 1968). Or, the experimenter provides key clauses to be combined into sentences (Opacic), 1973; Osgood & Bock, 1977) or completed to make sentences (MacKay, 1966). Or, the experimenter provides key sentences that must be used in a longer discourse (Levin, Baldwin, Gailwey, & Paivio, 1960). The sentence may be required in a particular place, e.g., at the end of the discourse (Osgood, & Bock, 1977; Tetroe, 1981).

    1.2.3 Recasting the presented message. Test subjects are asked to paraphrase presented sentences (Martin & Strange, 1968; Gleitman & Gleitman, 1970; Fillenbaum, 1971, 1974). A less controlled variant is retelling in one’s own words a whole story (Bartiett, 1932; Mandler & Johnson, 1977; Thorndyke, 1977; Mandler, 1978; Stein & Policastro, 1982; cf. surveys in Thorndyke & Yekovich, 1980; Beaugrande, 1982d). At present, these studies mainly concern comprehension and recall. The retelling as an act of production is less often in focus (e.g. Kintsch & Van Dijk, 1978; Beaugrande & Miller, 1980).

    1.2.4 Verifying a message. In the paradigm of “sentence verification, “ subjects must say if the statement conveyed by a sentence is “true” or “false” in everyday life (e.g., ‘Pines have purple needles’) or in a presented picture (e.g., ‘The star is above the plus’) (Trabasso, Rollins, & Shaughnessey, 1971; Clark & Chase, 1974; Carpenter & Just, 1975). Latency and error presumably reflect the processes of comprehending the sentence and then searching out and comparing relevant knowledge in memory (Clark & Clark, 1977: 111f).

    1.2.5 Responding to events or objects in a visual scene (cf. II.2.5).Test subjects describe in words what they see. The scene may be staged (Carroll, 1958; Levin, Silverman, & Ford, 1967; Osgood, 1971), or portrayed in pictures (Goldman-Eisler, 1961a; Fenz & Epstein 1962; Prentice, 1967; Learner Rommetveit, 1967; Tannenbaum & Williams, 1968; Flores D’Arcais, 1974; Kowal, O’Connell, & Sabin; 1975) or in films (Loftus, 1975). Many studies used “TAT” (“Thematic Apperception Test”) cards whose pictures are calculated to invoke particular topics (e.g. Benton, Hartman, & Sarason, 1955; Cervin, 1956; Sauer & Marcuse, 1957; Pope & Siegman, 1964; Martin & Strange, 1968). Most picture-response studies probed text production less than accompanying behavioral effects, {89} e.g., of audiences approving vs. disapproving (Cervin, 1956); of speakers knowing or not knowing that they’re being recorded (Sauer & Marcuse, 1957); of personal stress aroused by the topic (Fenz & Epstein, 1962); of using or not using key words (Martin & Strange, 1968); plus the physiological measures enumerated in III.1.4. More recent studies have probed how picture content is expressed in speech (Nelson, Reed, & MeEvoy, 1977) or writing (Mosenthal, Davidson-Mosenthal, & Krieger, 1981; Mosenthal & Na, 1981).

   1.2.6 Monologues. Test subjects receive a discourse topic and have to produce a monologue (Kanfer, 1959, 1960; Miller, Zavos, Vlandis, & Rosenbaum, 1961; Miller, 1964; Vlandis, 1964; Geer, 1966). These tests also focused on behavioral variables. Kanfer studied stress by applying electric shocks and clocking heartbeat and eye-blink. Miller, Vlandis, and their associates studied the impact of approving vs. disapproving audiences. Reece and Whitman (1962; cf. Reece, 1964) even bizarrely asked for monologues composed of disconnected words.1 [Evidently Reece preferred free associations (a common notion in the psychology of the times) over syntactic patterns because the latter present difficulties in deciding what counts as a single unit.]

  1.2.7 Dialogues. Test subjects are required to participate in a dialogue with the experimenter or with the latter’s confederates. The most common situation was the interview (Krause & Pilisuk, 1961; Allen, Wiens, Weitman, & Saslow, 1965; Kasl & Mahl, 1965; Pope & Siegman, 1965, 1966, 1968; Cassotta, Feldstein, & Jaffee, 1967; Cook, 1969). Control can be tightened by making subjects answer a list of questions (Turner & Rommetveit, 1967; Fillenbaum, 1968; Ervin-Tripp, 1970). Sometimes, the dialogue is staged in a therapy session (e.g. Mahl, 1956; Panek & Martin, 1959; Drennen & Wiggins, 1964; Pope & Siegman, 1966), especially if the experimenters specialize in treating mental patients. Once more, text production was chiefly a vehicle for studying stress, anxiety, and so forth. However, recent probes demonstrate how interviews help to explore discourse (Laboy & Fanshel, 1977; Wodak, 1981) and writing (Freedman, 1981; Odell, Goswami, & Herrington, 1982). 1.2.8 Self-commentary. Finally, text producers can be asked to report what they do (Lay & Paivio, 1969; Kowal et al., 1975; Daly, 1977; Flower & Hayes, 1980; Caccamise, 1981). Self-commentary suffers the limitations of introspective verbal reports (cf. Nisbett & Wilson, 1977, vs. Ericsson & Simon, 1980; Allport, 1980a).2 [Nisbett and Wilson argue that verbal reports are hard to disconfirm, and would be accurate only if “influential stimuli” are salient and plausible causes of “responses.” Ericsson and Simon reply that verbal reports would be reasonably accurate unless the verbalization directs attention to processes that wouldn’t normally receive it, so that people are inferring rather than remembering, and the reports are too general. The debate is inconclusive, since Nisbett and Wilson cling to behaviorist assumptions, while Ericsson and Simon adopt a cognitive approach. Many operations are too intricate and rapid to be consciously monitored and reported, e.g., memory searches for content {90} (cf. III.3. 10ff). Automatic processes are not open to conscious surveillance (III.1.21). The text producer might have an inaccurate impression, or want to make processing appear more orderly than it is (II.3.3). A logical analysis may miss the issues involved in an operational procedure (I.4.13). Producing the report itself adds another layer of organization to the experience (1.2.23.5). For all these reasons, self-commentaries can re-shape or distort the processes they recount.

  1.3 How far the findings from the experiments surveyed in III.1.2.1-7 apply to natural discourse is the main question for future research. At least three prospects seem plausible. Laboratory experiments might be: (a) less complex than everyday actions, because the experimenter requires simple, context-free tasks; (b) more complex than everyday actions, because people don’t have ready-made routines to apply; or (c) complex in a novel way not directly comparable to everyday actions.1 [Persons who act as experimental subjects may develop special skills for that purpose. Compare the differences in findings from psychology students vs. composition students on the same task in VI. 1.29f.] For example, repeating sentences is highly controlled, but also highly unnatural because of excluding normal decision and selection. Key words and paraphrases allow more variation, but are used more often in real discourse, e.g., in discussions or explanations. Monologues and dialogues are quite natural, but also quite complicated, even if the interviewer intervenes. There may be a frustrating trade-off: the more controlled the experiment, the less its findings can be generalized to natural discourse. An experiment that focuses on the syntax of isolated sentences and discounts the meaning and purpose of real discourse is likely to indicate a far more detailed and thorough processing of syntax than is done in normal communication (Johnson-Laird & Stevenson, 1970; cf. II.3.7, 12.2; V.2.16)

    1.4 A battery of physiological, psychological, and social evidence was gathered during (or soon before or after) the act of text production. Physiological measures included: blood pressure (Chappell, 1929; Innes, Millar, & Valentine, 1959); heartbeat (Kanfer, 1958); galvanic skin response (Chappell, 1929; Pope & Siegman, 1964);2 [“Galvanic skin response,” a measure of the fluctuation in the electric resistance of the skin, was used as an outward indicator of emotional events; like many such physiological measures (II.2.3ff), it doesn’t specify the details of the cognitive processes involved.] palm sweat (Kasl & Mahl, 1965); eyeblink rate (Kanfer, 1960); pupil dilation (Bernick & Oberlander, 1968); sensory deprivation (denying stimulation of the senses) (Suedfeld, Grissom, & Vernon, 1964); room temperature (Suedfeld, Vernon, Stubbs, & Karlins, 1965); occurrence of noises (like ‘uh’) in one’s speech (Mahl, 1956; cf. V3.12); and age (Kowal, et al., 1975). Psychological measures included: the “emotional responsivity scale” (Cervin, 1957); the “self-consciousness scale” (Levin et al., 1960); the “children’s test anxiety scale” (McCoy, 1965); and the “manifest anxiety scale” (Kasl & Mahl, 1965; Siegman & Pope, {91} 1965). Social measures included: isolation from other people (Suedfeld et al., 1964); audience sensitivity (Reynolds & Paivio, 1968); high or low exhibitionism (Levin et al., 1960; Paivio, 1965); warmth (friendliness, receptiveness, responsiveness) of audience (Reece & Whitman, 1962; Drennen & Wiggins, 1964; Allen et al., 1965; Siegman, 1968); adjustment to other people’s speech rhythms (Jaffee & Feldstein, 1970); deceitfulness (Mehrabian, 1971); and social class standing (Hawkins, 1973; Wodak, 1980).

    1.5 Such a battery of measures may seem impracticable, time-consuming, and exaggerated. However, text production is a complex activity sensitive to (and interacting with) a great variety of circumstances. We can’t tell a priori which physiological, psychological, or social factors might be relevant for language  performance. These experiments suggest that students should write a more stressful, disrupted, brief, and error-prone text if: (a) they write in favor of views they don’t believe in (I.3.16); (b) they are forced to work in isolation for a long time; (c) their teacher has a cold, unfriendly attitude; or (d) they come from working-class backgrounds. Further research on such questions is urgently needed.1 [In this perspective, Kerek, Daiker, and Morenberg’s (1980) sentence-cornbining project must be lauded for its thoroughness in controlling variables, even room temperature (cf. II.3.27).

    1.6 Because text production is so complex and context-sensitive, observed data are especially open to multiple interpretations (cf. I.1.5; Kowal et al., 1975; Flower & Hayes, 1981). Manifest events in spontaneous, natural text production are in principle hard to trace back to a single cause. The traditional solution was to “factor out” all causes but one or two and to assume that whatever happened was a simple result of manipulating “dependent variables” (in the sense of  III.1.2). Even if justifiable, this view leads to a fragmentation of theory and research: a host of cause-effect chains whose mutual functions within the overall system remain unclear. We would do better to incorporate multiple causes into elaborated models that stipulate their interaction. The effects of manipulating any single cause document a potential causality in natural production, though the cause of real events may still be undecidable. The relative probabilities of such potential causalities would best indicate the correspondence between complex process models and real-life activities.

    1.7 The DESIGN of process models is crucial precisely because of their intricate relationship to concrete events. This issue has not yet received the attention it merits. Much research uncritically accepted a model with an “encoder” sending a message to a “decoder.” These notions from engineering and information theory suited the physicalist outlook and the search for well-structuredness. Yet taken strictly, a “code” is a fixed set of unambiguous, arbitrarily defined symbols; and “encoding” is a mechanical interchange of symbols from one such set to another (e.g., Morse code  {92}  replacing letters with patterns of dots or dashes). No one has shown that the conceptual and planning systems in human communication qualify as “codes” in the same sense. Only the mapping between sounds and letters in spelling can fits the definition (Gough & Hillinger, 1980: 184), and only in part (cf. VI.31ff). The “encoding” model says nothing about the motives, decisions, and contexts of real communication (Shuy, 1981b).

    1.8 Explicit DESIGN CRITERIA can help to classify compare, or integrate possible models of communication and cognition. These criteria are to be operationally, not just logically, defined (I.4.13). UNIFORMITY vs. FREEDOM concerns the allowance a model makes for variations among individual people or processors. Processing can be well-structured without having to run the same way for everybody all the time. Since no one person possesses complete, infallible knowledge of a language, everybody builds a MODEL of it (III.1.28). Different people’s models agree enough for communication, and even for such tasks as revising someone else’s text (cf. III.1.26; V.3.4.8); yet vary enough to allow different skills and styles in speaking and writing. Hence, a theory should not define the norm so narrowly that numerous events appear deviant (cf.II.3.14; VI.3.22), nor so elastically that events appear accidental (I.1.2). Some investigators have eschewed the study of processes in fear that freedom might preclude generalization (I.1.3 — that each event might, at some degree of detail, be unique. However, the processes themselves must constitute a systemic and strategic basis for the “virtually unlimited flexibility, capacity for nuance, and creativity with which novel but appropriate behavior occurs as a function of an infinite number of contexts” (Spiro, 1977: 162; cf. I.1.7; I.4.8).

   1.9 REVERSIBILITY concerns whether the processes of production are the exact reverse (mirror images, so to speak) of those of reception (cf. II.2.22). This thesis would allow us to apply the more extensive research on reception to the less explored issues of production, e.g. reading vs. writing (Meyer, 1982; Shanklin, 1982) and spelling (V 1.30ff). Though logically plausible, reversibility is operationally doubtful. A receiver is partly recording and partly re-enacting the activities of the producer. The receiver can afford considerably more approximation, that is, can treat operations or materials in a provisional, fuzzy, or incomplete way, whereas the producer has to finalize things until the surface text can be executed. Production and reception approach the surface text from fundamentally different perspectives, each with its own implications for processing and decision-making. Hence, strictly reversible models are not likely to be confirmed by future research.

   1.10 SCALE designates the size of elements, ranging from LOCAL to GLOBAL (1.4.5). An operating system can enter either a local MICROSTATE or a global MACRO-STATE. A relation between micro-states yields a MICRO-STRUCTURE, and one between macro-states a MACROSTRUCTURE (cf. van Dijk, 1979). Upon entering a macro-state, the {93} processor expects certain classes of micro-states; conversely, a micro-state can be a clue about the macro-state being traversed. For example, the subject of a sentence requests the syntactic macro-state “noun phrase”; the microstate “definite article” is a clue that “noun-phrase” has been initiated (cf. demonstration in Beaugrande, 1980a: 44ff). On a larger scale, the noun phrase is a micro-state in the macro-state “sentence,” and the latter in turn a micro-state in the macro-state “paragraph,” and so on. As we easily see, a linguistic unit can be a micro-state for one process and a macro-state for another. The processor defines its scale by CHUNKING (treating several items as an operational unit, 1.4.5), and thereby determines COMPLEXITY (the configuration of part-whole relations, III.3.2.7). How much complexity is actually experienced need by no means agree with the complexity uncovered by an abstract analysis of structures (cf. II.3.12.2; 11.3.2iff, 37, 41). A skilled processor can use chunking to control and reduce complexity as needed (III.3.4.3).

    1.11 POWER designates the extent to which a type or definition is GENERAL (based on similarities) vs. SPECIFIC (based on differentiations). The more general, the higher the power (cf. Minsky & Papert, 1974: 59). For example, “event” is a more powerful concept than “motion,” and “object” more powerful than “part” (I.4.11.2). A processor can reduce its load by moving to higher power (III.3.4.4). Comprehension is eased by attending only to the overall gist of the text and discounting details (Kintsch, 1975; Schank, Liebowitz, & Birnbaum, 1978; Masson, 1979). Processing can be divided between ROUTINES applied to everyday requirements vs. SPECIALISTS that monitor and adapt on-line actions to fit unusual or variable conditions (cf. I.4.8; III.1.14, 27; VI.1.6).

    1.12 Power should not be confused with scale.1 [1 Units judged by their size (“ranks” in British research), e.g., word, sentence, and paragraph, have often been confusingly called “levels.” As noted in 1.4.14, a processing approach tends to cut the pie differently from abstract structural analysis. ] Local elements can be defined at either low or high power. For example, a single word might be defined at higher power as “modifier” or at lower power as “adverbial past participle.” If the surface text is damaged (garbled sounds, defaced print), the processor might assign only “word” and move on. Conversely, a global item like a book chapter might be labeled at high power: “Bloom goes home”; or at low power: “Bloom walks home through Dublin with Stephen Daedalus, climbs over the wall because the door key wasn’t found, lets Stephen in, lights a fire, makes cocoa,” etc.2 [This chapter of Joyce’s (1934: 650-721) Ulysses is in fact an experiment in the extremes of detail a narrative can attain, including a character’s unspoken thoughts, unperformed actions, and so on. Thus, scale and power are mae emphatically distinct.]

     1.13 DEPTH designates the LEVEL (in the sense of 1.4.2) on which processing is dominant (see III.2.5ff). A text can be processed on SHALLOW levels as a sequence of letters and words, e.g., by a young child; or on a DEEP LEVEL, e.g., as a hierarchy of ideas and goals. Deeper processing has more pronounced affects on memory and performance. Disambiguating words is deeper than checking their spelling (Bobrow & Bower, 1969). Finding anomalous meanings is deeper than watching for specified sounds or letters (Treisman & Tuxworth, 1974). Fitting words to context is deeper than finding rhymes for them (Craik & Tuiving, 1975). Making follow-up sentences is deeper than judging the meaningfulness of a sentence as it stands (Mistler-Lachman, 1974). These findings can be correlated with the scheme of “levels” going from the surface (the text as an artifact) to the “deepest” levels of main ideas and goals (III.2.3ff). The results of text processing can vary according to the dominant levels of operation.

    1.14 MEMORY CONTRIBUTIONS subsume whatever is put into the processing act from memory storage. A physicalist trend was to discount such contributions and depict processing as an ABSTRACTION OF TRACES from the input (the “stimulus”) (cf. Gomulicki, 1956; Broadbent, 1958; J. Gibson, 1966; E. Gibson, 1971). More recently, a CONSTRUCTIVE memory actively supplying world-knowledge has been affirmed (e.g. Neisser, 1967; Bransford, Barclay, & Franks, 1972; Bransford & Johnson, 1973; Ortony & Anderson, 1975). More recently still, researchers favor a RECONSTRUCTIVE memory that contributes both to processing the original events and to remembering them later (e.g. Spiro, 1977; Loftus, 1980; Rumelhart, 1980; cf. III.3.9); stored knowledge continues to evolve in memory as the person learns new things.1 [Evidence from word-association tests (cf. II.2.8) led to Tulving’s “encoding specificity” hypothesis: a reminder works only if it was stored at the time of the original experience (cf. Tulving & Thomson, 1973). This view, often contested as unduly restricting the adaptability of memory and interpretation (cf. Reder, Anderson, & Bjork, 1974), may be an artifact of odd laboratory tasks (pairing up words for a test). Keele (1973) suggests that virtually unlimited traces of experience may contact memory.] Construction and reconstruction are supported by: communities of routines and specialists; defaults and preferences; frames and schemas; and so on (I.4.8, 11.2; III.1.11).

     1.15 INHERITANCE is vital to a large system with a rich base of world-knowledge (cf. Hayes, 1977; Brachman, 1978; Fahlman, 1979). To save duplicating storage, a concept or operation INHERITS from a similar one by transferring specifications. An INSTANCE inherits from its CLASS (Walter Kintsch’s now-famous example was that Napoleon, being an instance of the class ‘people’, must have had toes). Or, a SUBCLASS inherits shared features from its SUPERCLASS (e.g., Frenchmen, Corsicans, etc. have toes because they are subclasses of the superclass ‘people’). CANCELLATION blocks inheritance by stipulating how a concept differs from its class or superclass (e.g., if we are told Admiral Nelson had only one eye). Inheritance can economize and raise power by replacing instances with classes, or classes with superclasses (e.g., by remembering ‘flowers’ after reading ‘tulips’) (cf. de Villiers, 1974; Frederiksen, 1975; Rumelhart, 1977a: 34; Kintsch & van Dijk, 1978); {95} details could be reconstructed if needed. ANALOGY permits inheritance among entities that are made comparable in special contexts (if we hear that Napoleon was a fox, we transfer traits like being quick and crafty); the interest and effectiveness of analogical language is shown by the enduring tradition of rhetorical “tropes” (metaphor, simile, metonymy, synecdoche, etc). Inheritance probably allows more economy than memory actually possesses. Strategic, frequently needed configurations are likely to be standing entries even if they duplicate others (cf. Kintsch, 1977: 290f). The waste in storage is traded for saving the time and effort that would be needed to assemble knowledge on demand (cf. III.3.9).

    1.16 DECOMPOSITION occurs when processing breaks elements down into components, i.e., lowers them in scale. The view that word meanings are made up of “minimal features” or “components” appeared variously in structuralism (Greimas, 1966), behaviorism (Osgood, 1963), mentalism (Katz & Fodor, 1963), and psycholinguistics (E. Clark, 1973; Gentner, 1978). “Each component,” according to Clark and Clark (1977: 509), “can be regarded as a procedure called up whenever that component forms the meaning of a word” (for example, ‘Napoleon’ might be broken down into ‘human’, ‘male’, ‘animate’, and so on). I see little evidence that decomposition is routinely performed during comprehension (Kintsch, 1974: 242; J. Anderson, 1976: 74; Hayes-Roth & Hayes-Roth, 1977). It appears to be a special case when some feature becomes relevant in context (Deese, 1971: 168; cf. Bolinger, 1965). Certainly, context is more essentially an act of assembling than one of disassembling (III.3.14).1 [Nearly all the “semantic feature” schemes look for them in an abstract catalog, like a dictionary run though an atom-smasher. This approach soon bogs down in the problem of how to invent and label the features. In a communicative context, there is no reason why a processor should be concerned with particular labels within a configuration whose meaning results from its coherence.]

    1.17 RESOURCE ALLOTMENT occurs when a processor distributes its capacities, such as memory, attention, and motor control, among the various demands of a system in operation (cf. III.3.2). Resources are nearly always limited, so that a TRADE-OFF between competing demands is frequently called for. Resources may be increased by special efforts (e.g. concentration, strain) or decreased by low motivation (e.g. boredom, lack of interest). Each process draws a LOAD on resources and becomes steadily more DOMINANT as its load becomes greater. The system tends to be heavily loaded by such factors as haste, distraction, emotional tension, uncertainty, and conflict among alternatives (Lashley, 1951: 11); or decision-making, trouble-shooting, novelty, danger, technical difficulty, and conflict with previous habits (Norman & Shallice, 1980: 21f). When the draw on resources becomes too great, the system enters a state of OVERLOAD that requires compensation. Unless resources are increased, operations will be DEGRADED by a decrease in completeness, consistency, or accuracy. {96} Skilled processing entails strategies to counteract overload and restrict degradation (III.3.4ff). For example, resources can be allotted to strategic CONTROL CENTERS within a configuration, and distributed from there (cf. III.2.14, 17; IV.2.6).

    1.18 Whether there is just one main supply of resources for all processes, or specific supplies for the various domains of cognition and action, is unclear (cf. Neisser, 1967; Kahneman, 1973; Allport, 1980b). Norman and Bobrow (1975) contrast “resource limitations” due to the processor against “data limitations” due to the materials to be processed. Granted that any person has a model (not the entirety) of language (III.1.8, 28), your limitations depend on your model. Practice would bring improvement if the model is well-designed and just needs rehearsing; but not if the design itself is unproductive, or unfit for the task at hand. The written modality at first confronts young children with data limitations. Once the new modality is properly acquired, resource limitations appear: how to use writing once you know the system. Finally, proficient writers could optimize their models until data limitations return: you can only write so fast within the mechanics of print, longhand, or type (III.1. 20; IV.1.12). Often, language development hits a BLOCK because one’s model is not able to evolve into a better approximation (cf. I.2.22, 23.3; I.3.6). Learners with so-called “literacy problems” are apparently locked into unworkable models (cf. I.2.22ff). The rote drills for beginning writers and readers may reinforce such models, rather than bringing about their reorganization (III. 1.28).

    1.19 If processing has WEAK SPOTS (characteristically troublesome operations), BOTTLENECKS (points where the quantity of materials needed is greater than can be accessed, stored, or transferred), and BLOCKS (points where operations can’t go on), the next question is where and why: in perceiving the world (Broadbent, 1958; Neisser, 1967); or in taking action upon what has been perceived and registered (Keele, 1973); and on so. Text production, especially writing, seems to operate near the threshold of overload (cf. Scardamalia, 1982), and to be beset by weak spots. In terms of problem-solving, (I.4.16), writing is hard when the problem space constantly shifts (Caccamise, 1981: 92), and the problem itself is uncertain, open only to “tentative solutions” (Odell, 1973: 42). Further obstacles include low knowledge about the topic domain (Voss, Vesonder, & Spilich, 1980; cf. I.2.8.8) and emotional stress (cf. Scardamalia, 1975). Revision can combat overload by focusing selectively on weak spots, and using resources from a longer time span (III.3.4.8f).

    1.20 Some resources are drained away simply by the mechanics of uttering or inscribing the surface text (cf. III.2.31; IV.1.11; V.1.35). Allan Newell (1973, cited in McCorduck, 1979: 265) remarks:

 

In spontaneous communication with speech, the human  appears not to be speech-limited, but rather thought-limited, whereas with writing the opposite is true. That is, a person knows what he wants to communicate faster than he can write it, but not faster than he can say it. Even when saying predigested material, our speech apparatus is never used at close to capacity.

 

The differences in rate and timing may affect the transition from speaking to writing (cf. II.2.29; II.3.38; III.3.35; V.3.31). The inscription action brings cognitive processing to bear on the surface text at a slower rate than does an articulation action (cf. IV.2.22). Also, the redundancies that fill out speech might carry over into writing (cf. V.3.10, 35-40).

    1.21 ATTENTIONAL processing conflicts with other operations at the same time, whereas AUTOMATIC processing does not (cf. Neisser, 1967; Kahneman, 1973; Keele, 1973; LaBerge & Samuels, 1974; Posner & Snyder, 1975; Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977; Allport, 1980b; Norman & Shallice, 1980).1 [Shiffrin & Schneider (1977: 155) define an “automatic process” as “a sequence of nodes that nearly always becomes active in response to a particular input configuration” coming either from outside or from inside the processor; “the sequence is activated without the necessity of active control or attention by the subject.”] Attentional actions tend to be serial, slow, easy to set up, alter, or suppress, and strongly sensitive to resource loads; automatic actions tend to be parallel, rapid, hard to set up, alter, or suppress, and insensitive to resource loads. Probably, expert performance relies on a strategic intersection and interaction of automatic processes, so that very complex activities can run without conscious supervision. On-line integration would constitute chunking: a stored package of such processes would constitute a schema (cf. Piaget, 1976; Norman & Shallice, 1980). Each mode of processing has its own dangers. Whereas attentional processes interfere with each other, automatic ones may cause trouble if they must be changed or prevented, or if their structures are similar enough to conflate (Norman, 1981). Moreover, a special processing obstacle (e.g., having to spell or pronounce a difficult word) could shift normally automatic processing to attentional, bringing a degradation in skill, fluency, or accuracy.

    1.22 Wherever text production is so complex that its processes cannot be done with attention, automatization is needed. For example, beginning writers must expend attention on the motor activities needed for “mechanics” (e.g., letter formation, spelling, or typing). This load on the system causes a shortage of resources elsewhere: the slow, laborious search and organization of content, including irrelevant associations; the deceptively numerous errors and inconsistencies; and so on (cf. III.3.32; IV.2.30; VI.35). Skilled writers handle mechanics automatically, keeping resources free for deeper, larger-scale concerns.

    1.23 SCHEDULING designates the order in which operations are carried out. An operation that can’t run without the results of another is CONTINGENT upon the latter, unless those results can be predicted. In {98} SERIAL processing, one operation at a time runs to its conclusion; in PARALLEL processing, various operations run concurrently and consult each other’s needs and outcomes. SCHEDULING ROUTINES would stipulate the usual order, while SCHEDULING SPECIALISTS would meet any special demands (cf. I.4. 8; III.1.11, 27). Since an automatic process can run at the same time with others, the transition from attentional to automatic processing demands rescheduling. A new skill could start in a SUCCESSIVE mode (serial processing), and evolve to a SIMULTANEOUS mode (parallel processing) (cf. Das, Kirby, & Jarman, 1979; cf. III.3.4.9). Since a new skill is continually in a process of experimenting with possible schedules, it would be prone to entail a high error rate (cf. I.2.17, 23.8; I.3.8, 11, 22, 24; VI.3.13).

    1.24 Several schemas could be active at once, if only one is dominant (III.1.17).1[Norman and Shallice (1980: 14) consider the selection of more than one “schema” (1.4.11.2) at a time “unlikely,” but they leave open the possibility of combining “two action sequences concurrently” into “a single, higher-order” schema. Thus, the “schema” itself can be created in context, not just looked up from storage (cf. also Rumelhart, 1980; Rumelhart & MeClelland, 1980).] Increasing skills can “consolidate schemas” (Piaget, 1976: 66), respecting their mutual contingencies. “Packages” of “sub-routines” can be set up to watch for their triggering conditions (III.1.21; III.31; 1V.2.30). Classic Gestalt theory stressed the “simultaneity of processing characterized by organizational laws stemming from the totality as such” (Piaget, 1976: 126ff). Luria’s scheme is similar (Das, Kirby, & Jarman, 1975: 82):

 

Simultaneous integration refers to the synthesis of separate elements into groups often taking on spatial overtones […] any portion of the result is at once surveyable without dependence upon its position in the whole […]Successive information processing [is] in a serial order […] the system is not totally surveyable at any point in time. Rather, a system of cues consecutively activates the components.

 

Luria classifies complex intellectual processes, plus perception and trace recovery of experience, as simultaneous; and syntax is taken to be successive. However, syntax is probably the outcome of transforming a simultaneous array into a linear sequence (11.3.16; IV 1.4, 7-9).

     1.25 THRESHOLDS are the criteria that INITIATE or TERMINATE a process under appropriate conditions. PATTERN-MATCHING is used to detect when conditions should trigger a threshold (cf. Morton, 1970, 1980), with a GOODNESS OF FIT ranging from exact to fuzzy (III.3.2.5). For high-speed operations without conscious supervision (“closed skills,” III.1.1), the setting of thresholds is the strongest means of control (Norman & Shallice, 1980).2 [In fact, Norman and Shallice (1980: 13ff) suggest that selection of a schema can be influenced only by setting thresholds; once initiated, the schema must run its course. Presumably, they have “closed” skills in mind here. “Open” skills could, for example, have such adaptive mechanisms as forming new schemas on the spot (see previous note).} The processor can consciously set an unusual value {99} (e.g., to suppress a habitual action by raising its threshold of initiation) that reverts to normal when supervision ceases (III.3.4.2). A strongly motivated writer, for example, sets high thresholds and requires exact matches, adapting as context evolves.

     1.26 Writing is plainly an open-ended task with no fixed point of termination           even though the concluding words get put on paper. The text could be always be reconsidered and revised (consolidated, developed, paraphrased, altered in voice, style, and register, etc.). The writer sets a threshold of termination that may be raised after the first draft is done, or may be attained only after several drafts. REVISION apparently entails a resetting of thresholds. A reviser who is not the original producer must re-enact production closely enough to build a model of the intended purpose and meaning and thus infer a strategic threshold (III.1 8) — further evidence that processing is well-structured, not chaotic (1.4.8).

    1.27 INTERACTIVE models have co-operating components, and MODULAR models have independent ones. The structuralist approach, with its isolated levels (II.1.15), and transformational grammar, with its groundings in axiomatic logic (II.3.12), favored modular theories in linguistics; more recently, interactive theories are gaining acceptance (III.2.1-5). However, this whole dichotomy fades if we postulate SPECIALIST COMMUNITIES that monitor ongoing processing and start or stop actions when their threshold conditions are met (cf. III.1.11, 25). Such communities are modules, yet interact wherever the system is motivated to use them (cf. Miller, Galanter, & Pribram, 1960; Newell & Simon, 1972; J. Anderson, 1976; Allport, 1980a). Past research tended to see entire levels as undifferentiated modules; and processing therefore appeared fragmented and diffuse (I.4.14; II.1.15). Power and efficiency would be much greater if similarly structured operations on various levels interact and merge (Woods, 1978), e.g. when word order conforms to the order of perception or memory retrieval (cf. II.2.23; III.3.19; 1V.2.56). Appropriate specialists could be set to detect these similarities and consolidate operations immediately.
     1.28 LEARNING occurs when a processor adapts and refines itself during either one operation or a series of operations (cf. II.2.6). Experiments often include “rehearsal” (intentional repetition to encourage learning, e.g., repeating a word to keep it in memory) (cf. Brown, 1958; Craik & Lockhart, 1972) and “distractors” (tasks that impede rehearsal or compete for resources) (cf. Peterson & Peterson, 1959). Whereas behaviorists believed that rehearsal alone causes learning by strengthening “stimulus-response” associations, it is now plain that the real benefits come from cognitive reorganization (Piaget, 1976, 1977; Greeno et al., 1978; Mandler, 1979). Apparently, one’s process model is impelled to evolve to a more efficient or inclusive version, such that the desired tasks become easier and more compact (cf. I.2.22; III.1.8, 18). For example, feedback can be used more precisely (III.3.2.2); operations can be shifted from attentional to automatic (III.1.21f); and so on. Skills remain OPEN as long as the model is evolving, and become CLOSED when the model becomes stable — thereafter, the rising curve of improvement levels out (becomes “asymptotic,” 1.4.15). Literacy education depends on finding tasks that motivate continual refinement of the learners’ language models. Activities like proofreading and editing will be helpful only if learners can relate shallow-level decisions and discrepancies to deeper-level ideas and goals (cf. III.3.7, 37; V.3.20, 22).

  1.29 A TYPOLOGY OF MATERIALS sorts out and classifies the artifacts upon which processing is carried out. For instance, reading and writing adapt to type of text involved. DESCRIPTION is centered on the properties and relationships of objects. NARRATION features temporal and causal sequences of events. EXPOSITION tells what something is or how it works. ARGUMENTATION tries to make an audience accept a thesis. SCIENTIFIC texts are intended to add to the general knowledge of a field; TECHNICAL texts present current knowledge to special audiences (e.g. technicians, operators, manufacturers); and DIDACTIC texts merely present it to an audience of basic learners. However, these types are only dominances, and actual samples are often complex mixtures of types (cf. Kinneavy, 1980). A typology of texts should be based on processes and contexts, not just on the features of artifacts (Gulich & Raible [Eds.], 1972; Meyer, 1975; Meyer, Rice, Knight, & Jensen, 1979; Beaugrande, 1981a, 1982d; Meyer & Freedle, in press).

   1.30 SIMULATION is the programming of a theory or model to run on a computer in imitation of real human processing. Successful simulation at least forces the investigator to state assumptions in an operational synthesis, and proves that a model is workable in principle1 [The “systemic grammar” of Michael Halliday (1969; cf. Berry, 1977) proved better for simulation than transformational grammar (Winograd, 1972; Davey, 1978), because the latter was harder to formulate as decision-making procedures. According to William Mann (personal communication), a new computer implementation of Halliday’s grammar is now advancing at the University of Southern California Information Sciences Institute.],  but not that humans act the same way. The human brain has large storage, but difficult access; the computer has instantaneous access, but limited storage (Loftus & Loftus, 1976: 128). The computer is also inept at recognizing useful, but approximative analogies and at commonsense reasoning in lack of knowledge (Collins & Quillian, 1972; Collins, 1978). Still, human brains and computers might ideally interact precisely where the one complements the other. AS the state of the art advances in computer technology, this possibility becomes increasingly attractive.

   1.31 The design criteria described in III.1.8-30are shown in Table l. These criteria can help to compare 

 able I. 

 

Table 1. Design criteria for process models

reversibility: production vs. reception

 scale: local vs. global

power: general vs. specific

depth: shallow vs. deep level

memory contributions: trace abstraction vs. construction vs. reconstruction

inheritance: class vs. instance; superclass vs. subclass; analogy vs. decomposition

resource allotment attention vs. automatization 

scheduling: serial vs. parallel;

successive vs. simultaneous

thresholds of initiation and termination

interaction vs. modularity

learning: open vs. closed

typology of materials

simulation 

 

various researchers’ approaches (Beaugrande, 1981a). Text production is so complex that the design of theoretical models is especially crucial. The latter in turn can provide the rationale for new methods in literacy education (cf. VI.3). Though we can’t predict  which models will gain acceptance, we can agree on some evaluative criteria that help to adjudicate the design of models, not just describe it. I suggest one set of evaluative criteria (listed in Table 2) by way of illustration (cf. Beaugrande, 1981b: 130f):

    

Table 2.Evaluative criteria for process models

_____________________________________________________________________________________

DIVERSIFICATION

CONSENSUS

ECONOMY

DYNAMICS

PLAUSIBILITY

COMPUTABILITY

INCREASING APPROXIMATION

ECOLOGICAL VALIDITY   v

_____________________________________________________________________________

 

1.31.1 DIVERSIFICATION: A model should unify the domains relevant to a given human aspect. Diversification retests the model’s plausibility with each added domain; supports clear antecedence relations among disciplines; keeps special cases from being mistaken for general ones; and combats reduction and fragmentation (cf. I. 1. 3, 12).

      1.31.2 CONSENSUS: A model should promote agreement among researchers by providing a clearly defined and consistently applied set of terms and concepts. This consensus would assist both integration of new findings from various directions, and comparison among alternative models.

     1.31.3 ECONOMY. A model should focus on those theoretical constructions

relevant for processing, rather than on all categories and features that could be invented (cf. II.1.6, III.1.16).

    1.31.4 DYNAMICS: A model should deal with processes and operations in natural contexts, not with static objects and artefacts in the abstract (I.1.3 ff).

    1.31.5 PLAUSIBILITY. A model should adduce realistic examples and respect the human resource limitations.

     1.31.6 COMPUTABILITY. A model should allow the statement of discrete procedures and steps that enable simulation (III.1.30).

     1.31.7 INCREASING APPROXIMATION: A model should be able to evolve toward a steadily more exact representation of the human aspects it addresses.

     1.31.8 ECOLOGICAL VALIDITY. A model should be an “account of how people interact with the ordinary world”; this priority does not mean “an end to laboratory experiments, but a commitment to variables that are ecologically important rather than those that are easily manageable” (Neisser, 1976: 7; cf. I.1.15; VI.3.22).

    1.32 These values did not always fare well in the three approaches reviewed in Chapter II. For example, diversification was done more through extrapolation than antecedence; consensus, economy, and dynamics were disregarded by structuralism and mentalism; and the other four standards fared badly in all three approaches. Today [i.e. 1984], research on cognition and communication is expanding so rapidly that it threatens to overwhelm the individual researcher An open discussion of model designs and of the ways to evaluate them could prevent a loss of orientation by reminding us of the larger issues at each step, and consolidating our advances as they occur. We stand the best chance of discovering facts if we inspect and critique the ways in which the design of our models influences what counts as facts, and for what purpose (I.1.4,7). 

2. THE PHASES OF TEXT PRODUCTION

     2.1 A model of text production should be situated within the general processing conditions outlined in III.1. Past research was simplified by relying on serial, modular models whose levels were processed one at a time (III.1.23). This scheme fit the analysis of linguistic artifacts into independent levels (II.1.15; II.3.7), as well as the design of psychological experiments measuring additive time (III.1.2; IV.1.8) (compare and contrast Chomsky, 1965; Lamb, 1966; Fromkin, 1971; Gibson, 1971; Shaffer, 1976; Jarvella, 1977). The popular, metaphoric “black box” model considered only the input and output of each stage, and disregarded internal operations, as illustrated in Figure 2..  

The impulse starts from “pragmatics” (purpose), passes through “semantics” (meaning), then “syntax” (phrasing), and pops out as “phonemics” (sounds) or “graphemics” (letters). Each box, however it may be labeled, carries out only the operations on its own level, and sends on final results. No box consults the others during operations. If failure occurs, the whole series would have to be traversed again from left to right

    2.2 The conventional proofs of experimental psychology depend heavily upon the notion of additive time (cf. Sternberg, 1969).1[On the controversial details of the “Sternberg paradigm,” see IV.1.8f.] in order to obtain quantitative results. The time on task is calculated to be the sum of the time spent on each component operation. Hence, proving a mental operation means triggering it with a suitable “stimulus” and showing a regular increase in performance time — what Posner (1978) calls “chronometric exploration of mind.” This proof assumes that (a) every mental process takes up real time (clock time), as compared to psychological time (the complex of overlapping, interactive processes); and (b) the experimenter really triggers the postulated operation, and it alone. These two assumptions are themselves debatable, but they are very hard to prove within a method that incorporates them directly into its proof procedures. Natural processing undoubtedly involves some operations, e.g., automatic processes that need not increase total time, because they run simultaneously with others.

    2.3 Recent trends favor parallel, interactive models (e.g. Marslen-Wilson, 1975; Norman & Rumelhart, 1975; Danks, 1977; Levy, 1977; Rumelhart, 1977b, 1981; Woods & Brachman, 1978a, 1978b; Flower & Hayes, 1980, 1981; Gould, 1980; Flood & Menyuk, 1981). Here, processes are able — though not required — to run at the same time and consult each other freely. In such models, serial actions, e.g., linearizing sounds, letters, and words, are the outcome of complex, concurrent processes (cf. Ch. IV). Linearity reflects the organization of the language modalities of speech and writing, rather than one-by-one mental processes. An elaborate parallel system can mimic a serial one if so required (cf. Anderson, 1976). Specialist processes monitor current conditions and become active in appropriate contexts (III.1.27).

    2.4 Interactive processing can streamline the reception and comprehension of texts on the way between the “surface” words or phrases to the “deeper” levels of main ideas and goals. Kintsch (1979: 326) points out that deeper processes can’t afford to wait for the shallower ones to finish. Drewnowski and Healy (1977) opine that parallel processing allows shallower levels to be shut down as soon as deeper levels terminate. Marcel (1980) proposes that processing is automatically and non-selectively carried to the deepest available level of processing (e.g., a word’s meaning could be recovered before its visual shape is fully identified). Such arguments should hold for text production as well: if reception is parallel, production could hardly be serial. But serial models are easy to design, whereas parallel, interactive models confront {105} the practising researcher with a host of difficult questions: (a) How are operations scheduled? (b) What is done simultaneously vs. successively? (c) Which operations are contingent upon others, and how strictly? (d) Which operations dominate others? (e) Which operations are, or can be, chunked? (f) How much is automatic, and how much is attentional? (g) How are thresholds of initiation and termination set or reset? (h) How are special problems resolved? (i) How many skills are open, and how many are closed? (j) How are limited resources allotted? (k) Where are the main control points in text production? (l) What factors contribute to overloading? (m) Does the production process have inherent weak spots or bottlenecks? (n) To what extent must processing adapt to each specific occasion? (o) How are processing strategies acquired, and how do they evolve? (p) What counts as a successful or satisfactory text? (q) How do skilled text producers revise? (r) How can text production be improved by educational training?

    2.5 In a parallel, interactive model, the language levels are processed in concurrent phases defined by their operations. These phases come and go as their thresholds of initiation and termination are met (III.1.25). Probably, the phases constitute functional units rather than temporal ones: the set of operations in a given phase would be done not all at once, but whenever conditions were right for triggering them (cf.III.2.25, 30, 32; III.3.9). Resource limitations should make it natural that processing dominance would be given to just one phase at a time, or to a few at most (III.1.17). Figure 3  illustrates the basic model, with the shallowest phase on top, and the deepest on the bottom. 17 The zig-zagged arrow suggests a gradual migration of dominance from deep to shallow during text production, yet with considerable freedom for strategic shifting up or down.

    2.6 The notion of “processing depth” 1[“The “deep/shallow” dimension must not be confused with alternative models on a “high/low” dimension: their “high” end is equivalent to the “deep” end here.] was originally advanced to account for differences in the impact of processing on memory and action (III.1.13). GOAL-PLANNING should thus be the deepest phase in the model because of its close ties to goals, desires, and emotions-states that are well remembered and strongly acted upon by most people (cf. Schweller, Brewer, & Dahl, 1977; D’Andrade, 1980). Goal-planning sets up the DISCOURSE GOAL (a future state of the world intended to be brought about via the discourse) plus the PLAN (the set of steps intended to lead to the goal) (1.4.11.3; cf. Larson, 1971). Multiple goals are ranked on a GOAL STACK and wait to assert themselves when conditions are right (cf. Rieger & London, 1977). Multiple plans can be held in reserve in case the first one fails (cf. Beaugrande, 1979d, 1980b, and references there). The participants MONITOR the progress of the discourse to update the status of their goals, and try to MANAGE it so as to attain the latter (1.4.11.5; 11.2.10).

     2.7 Membership in a culture entails approximative consensus about what goals are desirable because they affect one’s well-being (cf. classification in Beaugrande, 1980a: I 8 1). These desires may be physiological (e.g., health vs. illness, or satisfaction vs. deprivation); emotional (e.g., happiness vs. unhappiness, enjoyment vs. suffering, or stimulation vs. depression); social (e.g., acceptance vs. rejection, respect vs. disregard, or solidarity vs. confrontation); character-based (e.g., kindness vs. unkindness, modesty vs. vanity, or honesty vs. dishonesty); and so on. The usual desires are the basis for predicting or inferring other people’s goals from their texts and actions (cf. Schank & Abelson, 1977; Allen, 1979; Wilensky, 1980; Black, Wilkes, Gibbs, & Gibbs, 1982). A common default strategy is self-projection: assume other people have the same desires as you unless there is evidence to the contrary. Though an individual’s desires may deviate from those agreed upon by the society, the norm remains in force at least as a framework that makes us recognize such cases as deviant.

    2.8 The content (meaning) of discourse frequently depends on the presumed goals of the participants. This factor was treated in the literature on “speech acts” (cf. Austin, 1962; Searle, 1969, 1975; Cohen, 1978; Mann, 1980; Steinmann, 1982); and on “conversational implicatures” (Grice, 1975, 1978). Speech-act theory tends to treat language as sentences (Kinneavy, 1979a), due to the prevailing concerns of linguistics (cf. II.1.12; II.3.5, 13). Moreover, well-structured special cases with stable definitions (promising, threatening, thanking, greeting, etc.) have been favored over more common, but diffuse goals (pursuing social solidarity, making a good impression, demonstrating interest, passing the time of day, etc.). Politics could be defined as the domain in which communication and action are dominated by participants’ goals, if necessary at the expense of conventions such as conceptual meaning and consistency, factual truth, stable personal roles, and so on (cf. I.1.2; III.3.15, 30; VI.I.26). Discourse is more often “political” in this sense than most would  imagine.

    2.9 A broad theory of DISCOURSE ACTIONS should incorporate speech-act theory along with general aspects of goal-planning in discourse. An ACTION can be defined as an event brought about by an agent to change a situation in a way that wouldn’t have happened by itself (I.4.11.5), and a BLOCK as a state or event which impedes an action (III.1.20). Whatever bears on the action is termed RELEVANT. A DISCOURSE ACTION would be a discourse event performed to change a situation, and may be impeded by a DISCOURSE BLOCK, a well-known type being WRITER’S BLOCK (III.2.17, 20f; III.3.4; IV.2.50). The changed situation may contain the text producer’s GOAL STATE; or may be only one step on the way; or may prove irrelevant altogether. An INEFFECTIVE discourse action fails to bring the goal nearer. An INEFFICIENT discourse action wastes time and effort, e.g., in trial-and-error applied when goals and strategies are indistinct (cf. Kintsch, 1977: 441; 11.2.6, 9, 11). An INAPPROPRIATE discourse action is considered unfitting, whether or not it is {108} efficient or effective (e.g., a rude request is quick and gets results in some settings). The text producer can evoke cultural consensus to make the goal look worthwhile, e.g., politicians advertising their own goals as the ones any rational, forthright person should want.

    2.10 The goal-planning phase fits the text to the situation in several ways. First, the producer projects a ROLE, i.e. a pattern of actions and discourse expected from a participant. In “political” discourse, this role is carefully adjusted to encourage each audience to support one’s goals. Though consistency within a single situation and within all situations a person enters is considered normal, it is frequently subordinated to immediate goal-attainment. Second, the producer selects a TEXT TYPE that may be predominantly descriptive, narrative, expository, argumentative, and so on (III.1.29). Complex goals may call for a mixture of these types in a single discourse. Third, the PREPAREDNESS of the producer varies according to whether the discourse is important and anticipated. Everyday communication is often spontaneous (reacting to ongoing context) and extemporaneous (improvised) (III.1.1); these two conditions often coincide, but not always. (A practised orator who has a speech ready-made for the occasion speaks spontaneously, but not extemporaneously. A soap-box prophet of doom in a public park may speak extemporaneously, but, having no relevance to the real situation, not spontaneously.) Fourth, INFORMATIVITY varies between the discourse actions of INVOKING predominantly known content vs. INFORMING people of predominantly unknown or newly assembled content (I.4.11.7).

    2.11 Role, text type, preparedness, and informativity influence the producer’s choice of STYLE, operationally defined as a set of parameters that guide the selection among language alternatives (cf. VI.1.6). Audiences use stylistic “markers” (conspicuous choices, VI.I.4, 7) to classify roles and text types. The skilled producer has processing specialists for relating a variety of roles or types to appropriate styles (1.4.8; VI.1.6). Writing offers the advantage over speaking that a spontaneously produced original can be revised to contain consistent, appropriate style markers.

     2.12 The composition class is a special situation with elaborate presuppositions about purposes and goals. Writing (particularly in class) is mainly spontaneous, but not in the same sense as real-life discourse. Normally, the teacher is the audience, regardless of the intended addressee (cf. I.4.11.6). The main goal is to demonstrate text production itself, rather than to convey a relevant message (Applebee, 1982: 377; 1.2.13, 17). Since indistinct or unusual goals encourage wasteful trial-and-error, text production is likely to be inefficient and ineffective:

 

The freshman English theme is most frequently written without an explicit aim, takes no particular view of its subject matter, is oriented to no particular medium, and is preferably done with no serious thought preparetion. {109}In other words, it is aimless, modeless, mediumless, and unprepared. No serious professional writer would dream of producing a text under any of these conditions. (Kinneavy, 1979a: 13)

 

Larson (1971: 145) concurs that “students are insensitive to plans” and “unfamiliar with the notion of ‘purpose’ in writing”; they exhibit “abrupt changes in point of view, sudden changes in tone of voice, loss of direction midway through an essay, lack of transitions,” and “absence of convincing conclusions” —  common symptoms of trial-and-error. Students must perform under “unmotivating conditions messages that the writer has no impulse to send and that the reader (teacher) probably would not choose to read” without “being paid to be an examiner”; hence “inexperienced writers” will “think of purpose as what someone else wants of them” and will be “cut off from the impulse to say something, or from the sense that anything they might say is important to anyone else” in a “truly communicative situation” (Shaughnessy, 1977: 86). “Novice writers” “display an innocent lack of consideration for what their readers know and do not know, and for what they are or are not interested in” (Malmon, 1979: 364).

      2.13 This deep-level impasse can be resolved only by situating composition instruction in its broader social and psychological context. The sensitivity of text production to audience approval (III.1.4f) indicates that students’ texts will be non-fluent and error-prone if the teacher acts as an unfriendly, negative audience. Learners even imagine such an attitude without valid evidence. Amerindian children (Mesquaki Fox, an Algonquin people) construed the normal intonation of English teachers as “being mean” and “getting mad” (Coulthard, 1977: 49). Errors will increase if teachers are intransigent toward them, and decrease if discourse goals are realistic and clearly defined (cf. Williams, 1979a: 31) — not just to “inform,” “persuade, “ and so on (McCrimmon, 1976: 8). Students can gain some audience sensitivity and purpose by reading papers aloud to the class (Elbow, 1981: 96). By negotiating their own topics, students can assume a role as experts on matters about which they are knowledgeable (cf. I.3.24; III.2.19; VI.2.27).

    2.14 IDEATION is the next-deepest phase of text production and subsumes all activities that create an IDEA: a configuration of conceptual content that acts as a control center for building the text-world model (the total configuration of knowledge activated for processing the text, 1.4.11.2). A framework that regularly guides a person’s ideation can be termed an IDEOLOGY (1.3.16). Thus, “idea” and “ideology” are operationally defined as whatever assumes these control functions in context, rather than logically defined by degrees of abstraction. Wundt (1970 [19001) suggested that the components of an idea exist simultaneously (II.3.16) a view compatible with Gestalt theory (III.1.24). Or, the components of an idea might be successively assembled so rapidly and automatically that they seem to be {110} simultaneously present. According to Kintsch (1982: msp. 14f), memory search creates “ideas” for text production by setting up a “cue” that automatically fetches associated materials from memory storage — a case of spreading activation (I.4.11.2; III.2.20). Since automatic processing is not consciously supervised. this search can bring back irrelevant materials the writer should set aside or ignore (V1.2.26).1 [The findings of experiments on shadowing (III. 1.2. 1) that even conscious effort does not always enable people to ignore a message (Treisman, 1964) is probably an effect of automatic processing (cf. Posner, 1978: 96).] Selecting what you want to say from among all the materials activated by memory search can be an imposing task, the more so if the text topic is vaguely specified (III.2.18ff).

    2.15 Patterns of prior knowledge participate in ideation. A frame is an array of knowledge components generally associated in world knowledge; a schema is a progression of sequentially ordered components (I.4.11.2). However, these patterns would not normally be transferred intact over into text content, aside from the rare situation of saying or writing everything you can think of about a topic (cf. Charniak, 1972). Once automatic search has activated the whole pattern, the relevance of content must be judged (III.2.9; VI.2.2,11, 32). Often, the text producer has to search out and combine elements from several stored patterns and make a new unit. The proportion of this recombining determines the creativity of ideation (cf. I.3.7ff). On the audience’s side, the same proportion decides if the discourse action counts as invoking or as informing (cf. III.2.10).

     2.16 An idea might remain implicit, i.e., not be expressed (I.4.11.2; III.2.25, 35; V. 3.28); more often, it is made explicit (1.4.9) and thus becomes a TOPIC in the text. In a paragraph, a TOPIC SENTENCE can assume this function (IV.2.8; VI.2.14f, 17, 28). A topic which takes a stand on a problematic issue can be called a THESIS (VI.2. 1). In a series of texts within a discourse (e.g., a conversation or exchange of letters), a recurring topic would be the THEME. Making the idea the topic helps the audience re-enact the organization of content along the same lines used in production (cf. III.1.9). It helps to state the topic or thesis early in the text-the so-called “pyramid format” (Collins & Gentner, 1980: 61, cf. VI.1.15). The audience then has a clear control center for comprehension (compare Bransford & Johnson, 1973; Norman, 1973; Collins, Warnock, & Passafiume, 1974). Or, the producer may postpone the topic statement and invite the audience to infer it provisionally (V1.2.15). An idea that might seem unacceptable or incomprehensible to the audience could be saved until enough details have been presented to make it look reasonable and understandable (cf. VI.2.9f, 24).

     2.17 How ideation is organized and sequenced is thus a strategic issue for producer and audience alike. Both sides would fare best if the main idea came first, followed by the details of content (V1.2.1f, 7, 22). Processing would be roughly comparable on both sides-point of orientation first, {111} then elaboration or support — so that comprehension and learning would be efficient (cf. Horowitz & Newman, 1964; Meyer, Brandt, & Bluth, 1980). Starting content organization with the ideation phase is normal (cf. Grady, 1971: 348), but not compulsory. A producer might start with a flurry of details that only gradually coalesce into an idea. As long as no idea is acting as a control center, such discovery procedures would lack order and direction; writer’s block (III.2.9) could result either from not defining the idea (and bogging down in incoherent details) or else from not recovering relevant materials to develop the idea. Also, text production might be so complex the eventual topic is not the idea the producer started from; or, the latter may want to leave the idea implicit in order to conceal it. The audience may in turn infer an idea which is not the topic (cf. I.4.11.2), but which functions as a good control center for processing. Apparently, audiences are readily convinced by ideas they build themselves (cf. illustrations in Beaugrande & Dressler, 1981: 8, 154, 160, 176).

     2.18 Ideas can vary in scale and power (III.1.10ff). On the scale of dimension, Caccamise (1981: 36ff) found evidence of large “clusters” composed of overlapping “chunks” that in turn subsume groups of “idea units.” This hierarchy revealed how ideation differs according to specificity, familiarity, and writer’s age. On the power dimension, Christensen (1963, 1965) suggested that texts are organized in a hierarchy of generality via “co-ordination” vs. “subordination” (compare also Pitkin, 1969, 1977; Winterowd, 1970; Grady, 1971; D’Angelo, 1974; Nold & Davis, 1980;  and cf. I.3.18; VI.2.22).

     2.19 Like goal-planning, ideation occurs under atypical conditions in the composition class (cf. I.3.18; III.2.12f). In Kinneavy’s (1979a: 13) opinion, many students adopt “no particular view on the subject matter” and perform “no serious thought preparation.” They assume that their own views are not “possible content for academic statements” (Shaughnessy, 1977: 80). Applebee’s (1982: 375f) survey found most school writing to be simply an explicit rehearsal of materials not selected by the writer:

 

In the typical school situation the student’s body of relevant knowledge is completely circumscribed by the teacher’s knowledge of the subject area […] Marking practices tend to stress the accuracy of information presented, rather than the arguments that can be built up using that information [(cf. I.3.181)] […] We found many examples of essays that were little more than catalogs of “important facts,” related to one another primarily in the teacher’s vision.

 

Truncating the natural ideation phase leaves students ill-equipped to devise their own topics. They are unable to “gain access habitually to their own responses, their own thoughts” (Shaughnessy, 1977: 8 1). They may also fear their ideology could antagonize a teacher (1.3.16). The solution is often to select topics so banal they could offend no one — and interest no one. {112) Macrorie’s (1970: 12) samples of student “Engfish” markedly demonstrates the wishy-washiness that students assume as protective covering against the risks of cntroversy, individualism, and innovation (cf. I.3. 10; III.3.30f), e.g.:  

(28) I went downtown today for the first time. When I got there I was completely astonished by the hustle and bustle that was going on.

(29) It is hard to realize just how much you miss someone until you are away from this person. It seems that the time spent away from this person is wasted. You seem to wait and wait until you can see this person again.

 Personal experience is either watered down (28), or hidden behind the impersonal (29).

    2.20 CONCEPTUAL DEVELOPMENT is the phase in which ideas are enriched, elaborated, and integrated until the detailed text-world model (III.2.14) emerges. Here also, relevance screens out what is used from among what is available (cf. III.2.9, 15). And again, a writer can select either ready-made configurations, so that text figures in an invoking action, or creative, newly-built configurations, enabling an informing action (11.2. 10, 15). The mechanics of spreading activation (I.4.11.2) recover much trivial commonsense material even devotees of “Engfish” would find too trite to express. No freshman would venture to write (except in defiance):  

(30) “Downtown” isn’t really “down.” It’s the center of town, and has hills going up as well as down, as I learned when I went there today. It is made mostly of streets and sidewalks that help people get to stores, where they can buy the items they need to make their way through life from day to day.  

because every possible audience knows this only too well. Development must continue until less trivial content is recovered (V1.2.26ff, 32). Unfortunately, early results may promote writer’s block (cf. III.2.9, 17) by misleading inexperienced student writers to assume that all their content will be as trivial as the first things coming to mind. Teachers who assign a worn-out topic encourage this trap.

     2.21 Writer’s block also comes from DAYDREAMING: following up chains of association irrelevant to the task at hand. Spreading activation is largely automatic and runs without effort or control (III.2.14, 20; IV.2.30). Writing places a heavy load on processing (III.1.19f). Therefore, when the writer encounters overload, conceptual development tends to be dominated by spreading activation, and daydreaming ensues. The daydream provides a release valve for overload and degrades conceptual development by disregarding relevance (cf. III.2.15, 20). The content of a daydream usually comes from the associations which are best rehearsed or most salient to the writer at the time, e.g., recent experiences, personal well-being, desired goals, and so on. One of the trade-offs in text processing (cf. I.4.15) is between writing too slow vs. too fast. Slowing down may increase the danger of daydreaming; speeding up, as in “free writing” (V1.3.12), leaves little {113} time to develop content and screen for relevance. Thus, writing must be carefully scheduled (III.1.23), so that content is found and used without allowing the train of thought to wander off too far.

     2.22 The content of discourse is related to its REFERENCE in complex ways. Conventional treatments of reference are fond of using words or statements that match immediately visible objects (cf. II.2.5, 9, 11, 17; III.1.2.5; IV.2.56). However, discourse content typically invokes events and situations which are not immediately perceptible to the participants. What is referred to is decided by context: the text-world model is a system of relationships analogous (but not identical) to perceivable states of the experienced world (cf. 11.2.18, 23f) — i.e., organized in terms of time, space, causality, etc. Normally, processing terminates the regress to reality without going to verify the content of every text; situations calling for eyewitness evidence are comparatively rare. Thus, the issue of reference can’t be solved via philosophical speculation, but only by psychological discovery of how far people relate discourse content to real or imagined worlds in order to carry on satisfactory communication.

     2.23 Ideation and conceptual development have figured in composition courses as INVENTION (cf. VI.2.26, 32, 37). In the vitalistic, romantic outlook (I.2.23.7), invention seemed a miraculous, irrational inspiration from a muse; and composition teachers aren’t well armed to perform miracles (though often expected to). In fact, the semblance of irrationality was simply evoked by the recombining of old knowledge into new patterns (1.2.23.7; 111.3.15); true irrationality does not engender new order. Recently, the belief of classical rhetoricians that invention is teachable has been reaffirmed (e.g., Ohmann, 1964a; Gorrell, 1965; Rohman, 1965; Young, Becker, & Pike, 1970; Corbett, 1971; Larson, 1971; S. Miller, 1976; Young, 1976; D’Angelo, 1979a, 1979b; Kinneavy, 1980; Flower, 1981). Skilled text producers doubtless have acquired powerful strategies for discovering and organizing content across a wide range of knowledge domains — “prototypical thought patterns that transcend even the intellectual classifications of specific disciplines” (Shaughnessy, 1977: 257). Aristotle recommended using contrasts, conditionals, gradations, definitions, divisions, inductions, consequences, and so on (Cooper, 1932: 159ff). We need to explore and define invention precisely enough to be usable among contemporary students (cf. VI.2.6-37).

    2.24 Presumably, invention procedures guide MEMORY SEARCH. The strategies proposed by Young, Becker and Pike (1970), for instance, center on basic conceptual relationships: object/attribute, part/whole, class/instance, superclass/subclass, cause/effect, problem/solution, temporal sequence, and so on (cf. 1.4.11.2; VI.2.19-25). Students are encouraged to inquire how an object or event of the topic enters into these relations, e.g., what caused it and what results it will bring. Similarly, Burke’s “pentad” advises specifying “act,” “agent,” “agency,” “scene,”  and “purpose.” Such schemes should not overlook the danger that associative reasoning may be irrelevant or trivial in context (111.2.19ff. VI.2.26). Text producers should screen their materials to find what would excite INTEREST by being problematic, uncertain, or variable (cf. IV.2.11, 49; VI.2.8f, 32). Like goal-planning, ideation and conceptual development are instances of PROBLEM-SOLVING (cf. 1.4.16). The text producer finds and assembles content into an interesting configuration for the audience to re-assemble. Trivial content makes for trivial problems and elicits boredom; outlandish content, on the other hand, may make the audience’s problems too hard and elicit confusion (1.4.15).

    2.25 The EXPRESSION phase subsumes all the operations that map the conceptual configuration of underlying content onto language expressions (words, unitary phrases, etc.). Merely thinking of a concept might bring to mind one or more typical expressions for it (another instance of “spreading activation”). But the expression phase has to make all selections and close all remaining gaps not resolved this way. Hence, this phase is likely to be a distinct unit only in its functions: its component operations may run at diverse times. Some choices among alternatives may be quick and easy, and others protracted and difficult. A processor may have a mental image whose parts are not all equally easy to map onto language.1 [Apparently, mental imagery is correlated with language, but has its own processing conditions (Paivio 1975, 1978). The visual images of words would be an intermediate domain (cf Titchener, 1909; V.1.19f, 26, 33, 37, 48).] The continuity of experience and sensory impressions may be hard to render in discrete language expressions (cf. III.3.19).

    2.26 Theories of language have long been troubled by the controversy over how farr content (“thought”) and expression are identical or independent. To argue for total identity is to deny the possibility of full synonymity and translation. To argue for total independence is to ignore the speed and ease with which concepts activate expressions and vice-versa. My solution is to regard an expression as an item that activates an ordered set of hypotheses about processing actions on conceptual content in memory (cf. I.4.6). There is a straightforw