Chapter IV, Part 1

IV  

The Linearity of Text Production

1. TEMPORALITY VERSUS SPATIALITY

    1.1 The most essential precondition for the use of language elements is their multifarious combinability (Hartmann, 1963a, 1964), which renders them definable (cf. I.4.6, 10, 14; V.2.15, 17). Early theories of language looked at combination mainly as a word for word procedure. Linguistics looked for “immediate constituents,” “distributions,” “slots and fillers,” or “phrase structures,” to show how words were adjacent (II.1.12f; II.3.5). Psychologists postulated “associative chains” where each word is the stimulus for the next one (e.g. Washburn, 1916; cf. II.2.2). Statistical information theory (cf. Shannon & Weaver, 1949) computed the “transition probability” for any element by looking only at what element it immediately follows. The less often a particular transition between two elements occurs, the less probable it is and the more information it conveys.’ This theory was not intended to apply to language communication (cf. Shannon & Weaver, 1949: 8; Cherry, 1953: 383; Bar-Hillel, 1964: 222f). Unfortunately, its mathematical rigor and well-structuredness led to extrapolation in both linguistics (e.g. Lounsbury, 1954) and psychology (e.g. Goldman-Eisler, 1958a).

   1.2 We don’t need to think of language as processed word for word in order to relate linearity to probability. People are frequently able to guess the next letter or word in a sequence (cf. Shannon, 1951; Goldman-Eisler, 1958a, 1961a; Sprung, 1964; Stevens & Rumelhart, 1975). But this ability uses other cues (and other levels) besides the immediately preceding word. {148} To treat every word as a new decision point would entail an explosion of short-sighted calculations done right on the spur of the moment. People must apply contextual probabilities, not statistical ones. As discourse unfolds, people use all kinds of available cues to predict what will be said next. These predictions may not deal with the surface structure of the utterance itself, but with the general range of content that would fit an idea or topic (cf. Clark & Begun, 1968). Thus, the statistical probability of any sentence or utterance as such can be mathematically very low, yet its content and purpose could be highly predictable (cf. II.3.5). This approach allows people to overcome disturbances such as indistinct or missing words (III.2.38).¹ [Though widely used, the “cloze procedure” of filling in deleted words (Taylor, 1953) is not yet explained in terms of discourse processing. In particular, other insertions besides the original word need to be categorized, not just uniformly labeled as wrong. Words should be taken out not just at fixed intervals (usually every fifth word), but wherever specific hypotheses can be tested.]

    1.3 Word-for-word transitions could be assigned to a general strategy of ADJACENCY. “Put two related words next to each other whenever it’s convenient.” Adjacency contributes to the linearity of English texts, yet only as a preference frequently overridden in more complex and extensive frameworks. In an English “noun phrase,” for instance, only one dependent modifier can immediately precede the head; the others, though at a greater distance, are not necessarily less related to the noun (cf. IV.2.74). Conceptually related expressions will be recognized whether or not they appear next to each other in the sequence. In a passage like:  

(95) Am I a harp that the hand of the mighty may touch me, or a flute that his breath may pass through me? (Gibran, 1978: 7)  

the pattern of associations (‘harp-hand-touch’, ‘flute-breath-pass’) is clear despite their lack of adjacency in the surface text.

   1.4 Presumably, the linearity of language items is one aspect of the systematic linearity of human actions: The individual items of the temporal series do not in themselves have a temporal “valence” in their associative connections with other elements. The order is imposed by some other agent. This is true not only of language, but of all skilled movements or successions of movement. (Lashley, 1951: 116) The syntactic order of language would be a special case of a more general human faculty for linear activities (cf. Lashley, 1951; MacKay, 1974; McNeill, 1975; Piaget, 1976; Jaffee, 1977; Rumelhart & Norman, 1981). Within the range of “all cerebral activity,” “the syntax of the act” is “a generalized pattern or schema of integration which may be imposed upon a wide range and a wide variety of specific acts” (Lashley, 1951: 121f). If “intelligence” is the ability to fit one’s capacities to the current demands {149} (1.3.6), linear intelligence can designate all the abilities that enable linear processing of all kinds, including language.

    1.5 The text production model set forth in III.2.28-31 distributes linearization across two phases, one for phrases and one for sounds/letters. Logically, these two phases would duplicate each other, since the order of words entails that of word-components. But operationally, the sound/letter phase is responsible for the motor actions of uttering or inscribing (III.2.31). Decisions about phrasing are affected by deeper levels as well as by syntactic patterns. Errors in sound/letter placement are more common than errors in word order (cf. III.3.32; VI.16f, 42), apparently because the former level is processed in a more automatic, shallow-level mode than is the latter.

   1.6 The linearity of language is necessary because words have to be implemented one at a time. Spoken texts emphasize the temporal succession, whereas written texts emphasize the spatial one (cf. Nystrand, 1982c). The contrast results partly from the nature of the acoustic vs. visual modalities (cf. O’Connor & Hermelin, 1978). So far, most research has focused on temporality, because experimental results were so often measured by “reaction times” (III. 1.2; III.2.2). Spatiality and the relation of spatial to temporal processing were seldom explored outside Gestalt psychology (which was for a long time officially disdained by American psychologists). Still, we have some evidence that not all acoustic processing is temporal, nor is all visual processing spatial (Das et al., 1975: 82), not even in discourse. Hearers can discern different speakers by their spatial location without having to see them. Readers move through a text with characteristic timing as they fixate successive words with their eyes (Just & Carpenter, 1980).

    1.7 We need to sort out at least three senses in which discourse processing is “spatial.” In the first sense, “space” is a property of memory configurations that form arrays of concepts and expressions, not mere one-dimensional series (cf. II.3.16). Lashley (1951: 128) viewed “the translation from the spatial distribution of memory traces to temporal sequence” as the fundamental “problem of serial order”; “since memory traces […] persist simultaneously, it must be assumed that they are spatially differentiated.” Lashley envisioned this dimension in terms of the actual neurological structure of the brain and nervous system, but physiological space is not as decisive as “cognitive space.” An array can be simultaneously present in working memory, irrespective of participating brain locations. All the components of a chunk could be equally available at once (cf. III.3.5.2) — similar to the classic “Gestalt.” The search contaminations that lead to pre-activation and reactivation (III.3.32ff) would result from minor miscalculations among these components; if the chunk were already linear, such cases should be impossible. Similarly, problem-solving addresses a “problem space” containing the alternative states that might lead to the goal (cf. III.4.16). A chunk or a problem space no doubt contains linear chains too, but processing can access and survey them from a spatial perspective if needed. {150}

    1.8 At certain points of a discourse, the text producer selects a word or expression from a mental list of options (cf. IV.2.26ff). List-searching experiments raise several possibilities for deciding which word matches one’s specifications. Sternberg (1966) argued that the search through the list is serial and exhaustive, even if a match is found early. Other psychologists proposed that list-searching is parallel (Murdock, 1971), or serial, but not exhaustive, terminating when the item is found (Theios, 1973). Or, serial, exhaustive search might be suspended if list items are either highly familiar or highly unfamiliar — a view that fits the experimental findings best (Atkinson & Juola, 1974). “Reaction time” (III.1.2) does not always increase for longer lists, as would be predicted by the serial, exhaustive model advanced by Sternberg.

   1.9 In text production, the set of vocabulary words under consideration would seldom be as cleanly defined as the lists in the experiments just cited. Those lists were established at the outset of the experiment with items in an exact order, whereas a person’s vocabulary accrues from a rich, variegated experience in communication over a long time and is hard to keep in a set order. The “Sternberg paradigm,” and the “serial position effect” (that words in the middle of a list get mixed up more than those near the beginning and end) can hardly apply to text production, where vocabulary is large and diverse: active/passive, concrete/abstract, content words/function words, and so on. The implications of list-learning for text processing remains an open question for further research (cf. Kircher, 1971; Meyer, 1977; Brown, Campione, & Barclay, 1978; 11.2.8). Atkinson and Juola’s model could suggest why familiar items come to mind the soonest, and why word selection time need not vary according to the size of one’s vocabulary. But other factors can also affect the time spent on choosing one’s words (cf. IV.2.14-34).

    1.10 In the second sense, “space” is a perceptual category, i.e., one reflecting the dimensions of physical objects and events in perception. That the parts of a whole are simultaneously available to perception is clear from experiments on short-term sensory storage (cf. Sperling, 1960; Keele & Chase, 1967). Humans must be able to generalize from scenes they have experienced, or synthesize novel scenes by recombining traces from past experience. Humans can take successively presented visual arrays and unite them in memory to create a unified image (Eriksen & Collins, 1967; Kosslyn, 1978). Similar capacities could map a mental image onto a surface text, and vice-versa. One such tendency appears in the normal ordering strategies of discourse (III.3.19; IV.2.56). Mental images of words themselves appear to influence spelling (VI.20, 26, 33, 37, 48).

    1.11 In the third sense, “space” is the dimension in which the physical movements of a motor action are arranged (cf. Schmidt, 1975), such as uttering or inscribing. Speech phonetics has sought to explain “the well-known fact that acoustic correlates of a given phoneme, and therefore, {151} by inference, vocal tract configurations, are known to exhibit enormous variability, due particularly to variations in phonological context, speaking rate, and stress” (MacNeilage, 1970: 183). If we postulate a separate variant of each phoneme for every context (Wickelgren, 1969), we face a staggering repertory of over 100,000 items (MacNeilage, 1970: 185; cf. Denes, 1963). MacNeilage (1970: 185, 189) concludes instead that “the motor system is controlled by the results of an internal specification of certain spatial targets” whose nature is essentially dynamic: The direct control of speech production would originate with the reception by the space coordinate system of phonological information as to the utterance required. Then it would translate the information into a series of spatial target specifications. This would result in a series of demands on a motor system control mechanism to generate movement command patterns which would allow the articulators to reach the specified targets in the required order. Finally, these command patterns would be issued to the muscles. These operations resemble model-building, i.e., making a model of what is needed before trying to implement it.

    1.12 The same conclusions may be drawn for writing. Inscription is again “not a result of stored patterns of motor activity, but of the imposition of movement by a space coordinate system” (MacNeilage, 1970: 188). “Motor programs are flexible, interactive control structures, capable of calling upon sub-programs” and of “making local decisions as a result of current conditions” (Rumelhart & Norman, 1981: 7). If visual feedback about one’s hand motions is even briefly delayed, handwriting is severely degraded (van Bergeijk & David, 1959; Smith, McCrary, & Smith, 1960). A delay of 0.5 seconds makes handwriting almost illegible, and misspellings increase sharply, especially the doubling or tripling of letters. Figure 8 shows how two of my students wrote while looking off at a 90-degree angle toward the side opposite their writing hand (under each sample is an extract from one of their papers written at home). Vertical and horizontal displacement yielded uneven margins and crooked or overlapping lines. The students struggled to compensate for the evidently limited memory of line length and letter position. Note the erroneous, displaced attempt to cross, not dot, the first ‘i’ in ‘university’ in the top sample (unless the student mistakenly believed a ‘the’ to be there). Though these two students had fairly regular normal handwriting, less skilled ones could suffer overload from the complexity of motor actions in writing and perform below the language fluency they manifest in their speaking (cf. Connolly, 1982).

    1.13 The linearity of text production, especially of writing, deserves more thorough investigation in composition research. I have argued that this linearity depends on general human capacities for correlating time and space, i.e., for correlating the order of actions with their cognitive, perceptual, and physical dimensions. The processor stipulates the execution of the {152} surface text as a configuration of dynamic targets (IV.1.11) and monitors the actual outcome through open-loop and closed-loop feedback (III.3.2.2). Some margin of error is a side-product of the complexity and rapidity of the linearization. We need to explore writing skills in terms of how heavily this linearization loads the whole system of text production, and how large its {153} error potential is. Above all, we should describe how the linearity of texts is controlled by general principles that relate processing capacities and motivations to the surface text. This matter is treated in the next section.

2. SEVEN PRINCIPLES OF LINEARITY 

    2.1 A general theory of linear action should correlate the linear modalities of speech and writing with the levels and phases of processing. The goal-planning phase sets up pathways of actions that might lead to a goal (III.2.6ff). The ideation phase creates conceptual configurations that act as control centers for working with text content (III.2.14ff). The conceptual development phase enriches, specifies, and interrelates ideas (III.2.20ff). The expression phase assigns to concepts the appropriate natural language expressions (III.2.25ff). These phases do not immediately or directly yield the linearity of the surface text. Their patterns wouldn’t be the same on all levels, nor would they uniquely dictate the order of words or word components. The deeper phases include sets of options arranged in non-linear configurations, such as importance hierarchies, memory arrays, and lists of options (cf. III.3.7; IV.1.8ff). The two linearization phases for phrases and sounds/letters, in contrast, must conform to a sequential pattern that reflects both the input from the deeper phases and the concerns of syntax, utterance, and inscription (III.2.28ff).

    2.2 The LINEAR PRINCIPLES of so complex a processing system must be sufficiently powerful to work for all phases. Tasks whose linear output is similar in organization could be integrated and combined to run all at once (III.1.27). However, this tactic would be limited by asymmetry and redundancy among phases (III.2.27); and by the danger of confusing similarly structured actions (III.1.21). At any rate, non-trivial operations are probably needed to linearize anything but highly stereotyped texts.

   2.3 Seven general-purpose principles of linear operation appear necessary for any activity involving a linear modality (cf. Beaugrande, 1982f):

   2.3.1 The CORE-AND-ADJUNCT PRINCIPLE distinguishes between central and peripheral entities.

    2.3.2 The PAUSE PRINCIPLE allows the on-line sequence to be retarded or suspended.

    2.3.3 The LOOK-BACK PRINCIPLE subsumes all consultations of the prior discourse.

    2.3.4 The LOOK-AHEAD PRINCIPLE (the converse of look-back) subsumes all anticipations of the subsequent discourse.

   2.3.5 The HEAVINESS PRINCIPLE concerns gradations of importance, emphasis, focus, length, salience, or novelty, in the sense that these all draw a “heavier” load on processing (cf. III.1.17).

   2.3.6 The DISAMBIGUATION PRINCIPLE deals with excluding alternative patterns, both formal and conceptual.

    2.3.7 The LISTING PRINCIPLE handles the enumeration of comparable items in a sequence.

    2.4 These principles allow processing to navigate freely within an essentially linear modality such as language. We see their graphic representations in Figure 9.

The core-and-adjunct and heaviness principle reflect the functional equilibrium in the sequence: that some things have priority over others.¹ [On distinguishing between the core-adjunct and heaviness principles, see IV.2.53. The point is that any structure tends to have central vs. peripheral elements, irrespective of the processing load it imposes, either as a whole or via its components.] The pause principle regulates the rate and timing of the forward progression. The look-back and look-ahead principles access non-current stretches of the sequence. The disambiguation principle reduces branchings to one track, whereas the listing principle arranges branchings in a succession. These principles could also be stated in terms of problem-solving (cf. 1.4.16). The goal is the core, and the steps leading to it are the adjuncts; progress is gauged via look-back and look-ahead; pauses allow time to evaluate the situation and assemble resources; heaviness arises from the seriousness of the problem; disambiguation defines alternative plans; and listing enumerates multiple solutions to be tested.

    2.5 The seven principles are given operational designations (what they do), rather than logical-taxonomic ones (what they do it on) (cf. 1.4.14). A taxonomic or logical task-analysis could enumerate the structures of the various language levels, but not account for how those structures are set up and mapped among levels. This fragmentation would impede realistic models of cohesive and coherent discourse. Operationally, every linguistic rule or regularity is related (at least implicitly) to one or more of these linear principles. A position within the successive events of a communicative situation can even be occupied by a one-word utterance (sometimes called a “holophrase”) [The term “holophrase” reflects the supposition, aligned with transformational grammar, that one-word-utterances come from “underlying” sentences (cf. 11.3.13). But the evidence is against that assumption (Bloom, 1973).]. Thus, the linear principles seem to be necessary conditions for the application of any linguistic rule, but not yet sufficient ones. Each rule would require its own specifications (distinctions, categories, options, constraints, etc.) to meet the demands imposed on it.

    2.6 The CORE-AND-ADJUNCT PRINCIPLE distributes the flow of control within a structure. Processing would control the structure best by zeroing in on its CORE element, as if building a bridge: at strategic intervals, stable supports must be anchored before the complete superstructure is attempted. A language sequence often resembles a provisional multi-level bridge whose materials differ in their nature, span, and resiliency. The core- and-adjunct principle also regulates the complexity (part-whole relationships) of discourse processing (cf. 11.3.12f, 2iff, 36ff; 111.1.10-, 111.3.2.7). In the phase of sound/letter linearization, for example, the vowels appear to act as syllable cores, at least in English (V.1.39). In the phase of phrase linearization, content words would provide the most cores, whereas function words would routinely be adjuncts (cf. I.4.6; III.3.13; IV.2.11ff). Phrase-structuring would assign active status as a core or an adjunct to each “immediate constituent” (II.1.13). A core could be the head of a noun phrase or verb phrase, and modifiers would be adjuncts; on a larger scale, these roles could be assigned to the subject-verb unit vs. the rest of the clause; or to the main clause vs. all dependent clauses (cf. IV.2.12).

    2.7 In the ideation phase, cores would be the concepts higher up in a hierarchy of generality (V1.2.2, 22). In the conceptual development phase, the distinction between cores vs. adjuncts corresponds to that between primary vs. secondary concepts (I.4.11.2). Many of the secondary concepts are special cases of the primary ones (an “instrument” is an “object,” “motion” and “perception” are “actions,” etc.); and the secondary concepts are normally activated via some primary concept (“time” and “location” are applied to a “state,” “object,” “event,” etc.) (cf. Beaugrande, 1980a: 79ff). As in phrasing, the cores or adjuncts on the conceptual level can be defined or re-defined in the ongoing context. For example, a scientific treatise may use a “cause” as its primary concept, while the “event” being caused is well-known and hence secondary.

     2.8 This defining of conceptual status is conspicuously done via the topic, an idea made explicit (III.2.16). Topic concepts again resemble pilings for a bridge under construction (IV.2.6). Typically, the topic is announced early in the text, e.g., by a title or opening thesis statement (V1.2.2). Many paragraphs in written texts also begin with a “topic sentence” as a guide for processing the content (III.2.16; VI.2.14ff, 17, 28). However, topic announcements are essential only if the topic is shifting (V1.2.17), or else ambiguous (Bransford & Johnson, 1973). Otherwise, audiences can normally infer the topic from the core concepts around which they are organizing content. Text types have distinctive modes for such topic construction (III.1.29). The topic concepts for description are typically objects or situations; those for narration are typically events or actions arranged in a temporal and causal sequence to form a story line whose main core is the “turning point” where the protagonist’s goal becomes decisively attainable or unattainable (cf. Beaugrande & Colby, 1979).

    2.9 In the planning phase, the core would be the main goal, and the adjuncts would be the subgoals. Each goal or subgoal can be its own core for the plan states (e.g. preconditions) as adjuncts. In depth-first search, the planner tries to rush along the most direct path to the goal; in breadth-first search, the planner considers all candidate pathways before deciding which to follow (Winston, 1977). The first approach is apt when the problem is well-defined and its solution is obvious; the second approach fits better for vague, complex, or risky problems (cf. Beaugrande, 1979d, 1980b). To persuade an unfamiliar audience, for example, you may have to try a range of methods, monitoring the reactions for signs of success or failure.  

 2.9 In the planning phase, the core would be the main goal, and the adjuncts would be the subgoals. Each goal or subgoal can be its own core for the plan states (e.g. preconditions) as adjuncts. In depth-first search, the planner tries to rush along the most direct path to the goal; in breadth-first search, the planner considers all candidate pathways before deciding which to follow (Winston, 1977). The first approach is apt when the problem is well-defined and its solution is obvious; the second approach fits better for vague, complex, or risky problems (cf. Beaugrande, 1979d, 1980b). To persuade an unfamiliar audience, for example, you may have to try a range of methods, monitoring the reactions for signs of success or failure.

      2.10 The order in which cores or adjuncts are selected or presented is a difficult issue. If the bridge analogy (IV2.6, 8) holds, the core should be activated before its adjuncts, though it may not be immediately implemented in the text. However, there is evidence from both speech and writing that this order doesn’t always happen. In the phrasing phase, adjuncts such as articles and prepositions are sometimes uttered or written down before {157} their head nouns are chosen (III.2.30; IV2.27ff). Possibly, the category of the core is set, but not the specific item to represent the category. For instance, writing down a preposition or a determiner commits the text producer to the category “head noun,” even if no particular noun has been selected.

    2.11 Efficiency would be greatest if conscious, attentional processing could focus on the core, whereas adjuncts should be processed automatically (cf. III.1.21). Such appears to be the normal case in the treatment of content vs. function words (cf. IV.2.6). If content words have denser search grids, as argued in III.3.13, their recovery would naturally demand more resources. This factor might disrupt the linearization of phrasing in several ways. If content words or their concepts are complex, unfamiliar, specialized, or creative, memory searches could be taxing and time-consuming enough to block the normal core-first method. The text producer might decide on a powerful category for the core element and set the search running; meanwhile, less problematic adjuncts could be chosen and implemented, such as articles and prepositions. The core search might terminate and bring the needed item (a) in time to say or write it at its proper position; (b) after an additional pause (IV.2.28ff); or (c) at a point beyond the proper position (post-activation, cf. III.2.30; III.3.32f). In contrast, a core element presenting no search problems could be selected before any of its adjuncts, so that phrasing would have all its materials ready at hand.

    2.12 This account indicates that the actions of constructing clauses and sentences can be affected by factors not usually considered “syntactic”. In an English declarative sentence, the preference order would be to have the first noun phrase (usually determiner + modifier(s) + head) as subject, and the following verb phrase (usually verb + modifier(s), or auxiliary + verb + modifier(s)) as the predicate. Within each of these constituents, the core is normally the noun or verb, while modifiers and function words are adjuncts. All parts of the constituent should be placed in one of the buffers of working memory (III.2.29; III.3.5.4). The buffer should hold them long enough to label them and create the instructions that will output them into a proper phrase. Whether a text producer uses simple or complex phrasing would depend on how well working memory masters these tasks. If the materials are difficult to find and arrange, the buffer would tend to hold smaller amounts and phrasing would be simple. If the materials are familiar and easy to work with, the buffer could hold more, and complex dependencies would be easier to formulate. If this account holds, then syntactic complexity could hardly be a stable trait of a person or an age group (cf. II.3.23).

    2.13 The recursion of cores or adjuncts might also be controlled by buffers. In contemporary English, recursion applies mostly to content words: two or three nouns, verbs, or modifiers in a row are vastly more common than two or three determiners (articles, deictics) prepositions, or conjunctions in a row. Intriguingly, function-word recursion is a common {158} writing error, e.g., ‘the the’, or ‘and and’ — just what would be predicted if these word categories receive little attention (cf. V.1.46). If recursion requires elements that are both conceptually and syntactically comparable (IV.2.71; IV.3.39), then one search grid could gather several such elements into a buffer to await output. However, if content words have dense search grids (III.3.13), a long recursion might hinder easy sorting within the buffer. Hence, the fact that more than three elements are not common in recursions would make good sense. Special effects (surprise, confusion) would result from exceeding this strategic length, e.g., in this passage with nine present participles deliberately exaggerating the foolish excess of ‘Protestant taboos’:  

(96) the entire cupboard of Protestant taboos against drinking, lusting, gambling, staying out late, getting up late, loafing, idling, lollygagging around the streets, and wearing Capri pants (Wolfe, 1977: 17)  

For the same reason, recursion of highly complex elements would be hard to manage. For instance, a series of more than three dependent clauses in one sentence is much rarer than a sequence of more than three modifiers.

    2.14 The PAUSE PRINCIPLE meets the need to retard or suspend the linear sequence from time to time,¹ [For this discussion, the distinction between “hesitations” and “pauses” will not be maintained; but some of the literature cited in IV.2.19 does so, particularly following Mahl (1956) and Mactay & Osgood (1959). See Rochester (1973) for a fine survey.], for example, to bring the processing of the current stretch of text up to a reasonable threshold of termination (cf. III.1.25); to plan out the next stretch; and so on. Pauses and hesitations, being observable and measurable, have figured in much psychological and linguistic experimentation (cf. surveys in Rochester, 1973; Rosenberg, 1977b; Matsuhashi, 1982). In the behaviorist outlook, reaction time or latency (time between receiving a stimulus and starting to perform) and time on task (time between start and finish) were standard indicators of the nature and difficulty of an operation (III.1.2). Fluency (the rate and the smoothness of transitions in a task) was the principle indicator of learning (II.2.5). If these notions are to help us interpret pauses in an activity as complex and elaborate as text production, observed data should be carefully fitted to correspondingly complex and elaborate models of the production processes.

    2.15 A word-for-word model, e.g., one conceived in terms of word chains and transitions (cf. IV.1.1f), would predict pauses and hesitations failing before any unusual or difficult word choice. Goldman-Eisler (1958b: 67) started from the behaviorist view (cf. Hull, 1943) that conditioned habits control what is probable:

Fluent speech was shown to consist of habitual combinations of words such as were shared by the language community and such as had become more or less automatic. Where a sequence ceased to be a matter of common conditioning or learning, where a speaker’s choice was highly individual and unexpected, on the other hand, speech was hesitant.

Maclay and Osgood found that pauses (either silent or “filled” with nonword sounds like ‘urn’ or ‘erm’) appeared more often before content words than before function words-the latter class being of course smaller to choose from than the former. Restarts (going back and recommencing a stretch, cf. IV.2.50; V.2.34; V3.20f, 33) included any already uttered function words at the start of a phrase. If restarts included revisions, content words were more often changed. Thus, the customary distinction between the two word types was indeed reflected in processing (cf. III.3.13; IV.2.6, 11f). These findings suggest that phrasings are at least sometimes selected before the expressions that go into them (cf. II.2.14; III.2.30; IV.2.10, 27). Boomer (1965) reached a similar view by analysing pauses not in terms of the sentence, but in terms of the utterance unit he called “the phonemic clause”: a segment of speech having only one primary stress and ending with a terminal juncture (Trager & Smith, 1951; compare the “tone group” in III. 2.28). In extemporaneous speaking on such everyday topics as “hobbies, sports, summer vacations” (1965: 150), Boomer found that pauses did not routinely fall before specific rare or difficult word choices, but rather after the first word (usually a function word) of a phrase or clause. His conclusion (1965: 156) would be compatible with the notions of probabilities and transitions (cf. IV.1.1f):

The initial word in a phonemic clause sets certain constraints for the structure of what is to follow. The selection of a first word has in greater or lesser degree committed the speaker to a particular construction or at least a set of alternative constructions, and has also foreclosed the possibility of other constructions […] Hesitations in a phonemic clause are most likely to occur after at least a preliminary decision has been made concerning its structure and before the lexical choices have been finally made.

Here also, memory search would be a key factor (cf. IV.2.26ff).

    2.16 If clauses and sentences are the basic units of language processing, a text producer should pause at the end of each clause or sentence, plan out the next, run straight through, and pause again. Goldman-Eisler (1972: 103) hoped this hypothesis would relate grammar to “objective behavior and psychological reality” (cf. V.2.7-17) (perhaps resolving the antagonism between behaviorist and mentalist viewpoints?):

 if in the flow of spontaneous speech the transitions between its various constituents between words, clauses, co-ordinate and subordinate […] are of characteristic duration, then we can speak of their differential psychological reality and draw conclusions as to the degree of integration and independence of any of these units.

She found that the longest pauses appeared between sentences or before coordinated clauses. Pauses got shorter in this progression: (a) before {180} non-relative dependent clauses, (b) before relative dependent clauses (especially beginning with ‘which’), and (c) within clauses. She concluded that “the grammatical description of each of the clause types has its quantitative reflection in the length or absence of pauses which precede it in natural unprepared, and spontaneous speech” (1972: 106f). To some extent, the word-for-word transition model in her earlier work (1958a) was traded in for a clause-for-clause transition model.

    2.17 However, two major difficulties arise from Goldman-Eisler’s studies (cf.   critique in Boomer, 1965: 149; Rochester, 1973: 56f; O’Connell, 1977: 310). The first is the biased sample quite unlike “natural, unprepared, and spontaneous speech” (IV.2.16). Such speech is frequently hard to identify as a sequence of clearly marked sentences (II.3.13; V.2.18; V.3.7ff). Even many established, prolific scholars produce irreduceably ungrammatical sequences in extemporaneous discourse, e.g. (cited also in I.4.11.2):  

(3) As far as I know / no one has done the / in a way obvious now and interesting problem of // doing a / in a sense frequency study of the alternative /1 syntactical / uh / in a given language / say like English / the alternative /1 uh I possible structures (from Maclay & Osgood, 1959: 25)  

In early work, Goldman-Eisler (1958a: 99) “eliminated a large mass of recorded material of spontaneous speech,” since “even with highly educated speakers, spontaneous speech is such that well constructed, grammatically correct sentences spoken without repetition or midway changing of grammatical construction, etc., are few and far between.” Later (1972), she picked out skilled “academics” (including a radio broadcaster) whose speech was fluent and grammatical, the more so as literate speakers may tailor the habits of their speech to those of their writing (V.3.27). She increased regularity even more by allowing one group to study written texts silently and then read them aloud. In addition, a single secretary made the transcripts, imposing interpretation and punctuation. Thus, speech sample was filtered through the influence of writing in three ways.

    2.18 The second difficulty is that pausing may reflect several motives at once, not just the formatting of sentences and clauses (cf. Kowal, O’Connell, & Sabin, 1975; Flower & Hayes, 1981; Matsuhashi, 1982). To sort out the possibilities, Goldman-Eisler (1972: 110) distinguished between “cognitive pauses” for planning and selecting materials, vs. “rhetorical pauses” for demarcating sentences and clauses to guide the audience. Reading aloud ought to eliminate the need for the “cognitive” type (cf. IV3.10). Yet the findings on “miscues” in normal reading (cf. IV.2.48) suggest that reading aloud is by no means trouble-free, even if materials are familiar. Another line of Goldman-Eisler’s (1972: 111) argument was that “in most cases, a sentence presents the externalization of a thought unit” (cf. II.3.42; Chafe, 1982), so that her findings should apply to conceptual organization as well as grammatical. Yet she did not explain what “thought units” are “externalised” {181} by her clause types and how. And her argument is circular insofar as she based her definition of a “thought unit” on the sentence itself (O’Connell, 1977: 310). Taylor’s (1969) study found, on the contrary, that pausing varied according to concreteness vs. abstractness of topic words, whereas length, type, and syntactic complexity of sentences made no difference. Finally, no provision is made for pausing to look back rather than ahead (cf. IV.2.31; IV.3.11).

    2.19 Experimental findings indicate how unlikely it is that a single cause could be responsible for all the pauses and hesitations in text production. All of the following have been pursued as plausible causes:

2.19.1 type of text or discourse (especially description vs. explanation) (Goldman-Eisler, 1961a, 1961b; H. Levin, Silverman, & Ford, 1967; Lay & Paivio, 1970; Matsuhashi, 1981, 1982);

2.19.2 degree of conceptual integration (Caccamise, 1981; Spittle & Matsuhashi, 1981; Chafe, 1982; Cooper & Matsuhashi, 1982; Matsuhashi, 1982);

2.19.3 concrete vs. abstract topic concepts (Reynolds & Paivio, 1968; Taylor, 1969; Matsuhashi, 1981);

2.19.4 textual ambiguities (MacKay, 1966; Siegman & Pope, 1966);

2.19.5 reductions of coherence (O’Connell, Kowal, & Hömann, 1969);

2.19.6 breathing patterns (Barik, 1968; Hixon, Mead, & Goldman, 1976);

2.19.7 personal rates of speech (Maclay & Osgood, 1959; Pope & Slegman, 1968; Jaffee & Breskin, 1970) and of writing (Matsuhashi. 1979; Flower & Hayes, 1981);

 2.19.8 extent of automatic processing (Butterworth, 1975), as contrasted with creating new patterns (Cooper & Matsuhashi, 1982);

2.19.9 stages or cycles within a conversation (Henderson, Goldman-Eisler, & Skarbek, 1965; Butterworth, 1975; cf. VI.2.3-5);

2.19.10 length of the text (Feldstein, 1962; H. Levin & Silverman, 1965);

2.19.11 difficulty of speaking task (Goldman-Eisler, 1961a; Lay & Paivio, 1969);

2.19.12 speaking vs. reading aloud (Henderson, Goldman-Eisler, & Skarbek, 1965);

2.19.13 anxiety (Mahl, 1956; Krause & Pilisuk, 1961; Kasl & Mahl, 1965; Siegman & Pope, 1965 ; cf. III.1.4);

2.19.14 stress (Cervin, 1956; Casotta, Feldstein, & Jaffee, 1967; Lalljee & Cook, 1967; H. Levin & Silverman, 1967);

2.19.15 adjustment to another speaker (Jaffee & Feldstein, 1970);

2.19.16 speaking alone vs. to an audience (H. Levin & Silverman, 1965);

2.19.17 visible vs. invisible audience (Feldstein, Silverman, & Ford, 1967);

2.19.18 face-to-face vs. telephone conversation (Kasl & Mahl, 1965);

2.19.19 approving vs. disapproving audience (Cervin 1956; Pope & Siegman, 1966; Murray, 1971);

 2.19.20 likelihood of being interrupted (Lalljec & Cook, 1967);

2.19.21 exhibitionism of speaker (Levin, Baldwin, Gallwey, & Paivio, 1960; Paivio, 1965);

2.19.22 deceitfulness of speaker (Mehrabian, 1971):

2.19.23 age of speaker (Kowal, O’Connell, & Sabin, 1975);

2.19.24 social class standing of speaker (Hawkins,1973;Wodak,1980).

     2.20 These factors are all experimentally documented. Others might well be envisioned, such as: mental updating of a scene or event being described (IV.2.33); the social standing of one’s audience; the value of the goal, and one’s hopes of attaining it; one’s enthusiasm about an original idea; the social implications of literate activities; etc. Just the listing in 2.19 reveals how observed data, such as a pause or hesitation, may relate to an interaction among underlying causes, or to an interaction of these. Manipulation of one variable may affect others in complicated, unforeseen ways. For example, if one’s audience shows disapproval, the text producer may become anxious and consider the situation stressful (IV.2.19, 13, 14) — three reasons why speech would be hesitant and disjointed, and the resulting text brief and poorly integrated (III.1.5).

    2.21 We can thus appreciate the acute dilemma of complex process theories: that observed data can in principle be interpreted in many ways, depending on one’s model (I.1.5; III.1.6). We can manipulate any of the factors in IV.2.19, and no doubt others as well, to influence pause ratios. The tighter the control on performance, the more secure the cause-effect relationship will be. For instance, Lackner and Levine (1975: 109) had people study and memorize sentences on cards for days before the laboratory session. Lindsley (1975: 4) had people practice choosing from a list of four subjects to use in a predetermined sentence form. But such tight controls suppress normal search and decision-making operations and thereby distort our view of natural discourse processing (cf. III.1.3). The cause of an observed pause in spontaneous communication cannot usually be determined with certainty, but at most with fair probability. Similarly, the evaluation of text production can hardly be based on observed pauses alone, since the  factors responsible may be either positive (e.g. careful selection) or negative (e.g. anxiety) (cf. II.2.28f; IV.2.29, 34). In a general way, though, educational training could be designed to strive for conditions that don’t aggravate the potential disfluency of text production (IV.2.34).

    2.22 Most pause studies examined speech. It would be useful to see if comparable patterns also appear in writing (cf. Matsuhashi, 1979; 1981, 1982; Flower & Hayes, 1981; Spittle & Matsuhashi, 1981; Cooper & Matsuhashi, 1982). On the one hand, these patterns should be different because writing (a) runs at a much slower rate (Horowitz & Berkowitz, 1964; {163} Horowitz  & Newman, 1964; Blass & Siegman, 1975; Gould & Boies, 1978b; Gould, 1979; Chafe, 1982); (b) does not immediately interact with other people’s text production (cf. Rosenberg, 1977b); (c) is often not subject to interruption (cf. Lalljee & Cook, 1967); and (d) has no motive for inserting “ rhetorical pauses” to signal sentence or clause boundaries (IV.2.18; cf. Matsuhashi, 1982), since the audience is not present. On the other hand, patterns might be similar because writing (a) shares with speech the system of text production; (b) is learned later in life than speaking (V3.3); and (c) entails sounds, not just letters, in spelling and word identification (V.1.31ff). Thus, learning to write involves both the extension and the reorganization of skills already well practised in speaking (V.3).

    2.23 To observe patterns of action, I set up a television camera to film people as they wrote (cf. Matsuhashi & Cooper, 1978; Matsuhashi, 1979). When they finished, they spoke the same messages they had written, though without specially trying to reproduce the exact words; this performance was taped and transcribed. I sifted through the films to see where writers paused, crossed out, went back, or otherwise departed from the left-to-right sequence. Among the thirty-two writers, only two wrote without any significant occurrences of these types: a professor of English and an undergraduate student in broadcasting, both writing on familiar topics. Everyone else, myself included, paused here and there (not just between sentences); many of us crossed out words (either immediately or later), or went back to make insertions.

     2.24 Twenty of the test persons came from composition classes. They were assigned a simple, well-structured task: to give discursive directions (no maps) on how to get from the north-bound Interstate 75 west of Gainesville to their own residence, mostly in the campus dorms. If they had no cars and were unfamiliar with the town, they described how to get across the campus on foot. The task has a clear beginning and end, and the ordering of content closely follows the spatial progress of an imagined person (cf. IV.I.10). However, problems can still arise. With very few exceptions, Gainesville’s streets are not named, but numbered, moving out in the four directions of the compass from the central axis of Main Street and University Avenue. Street designations are hard to remember: a number, a direction (NW, NE, SW, or SE), and a type (Street, Avenue, Place, Lane, Road, Terrace, Court, etc.), e.g. ‘2306 SW 13th St.’. Besides, many people have only vague mental maps of their towns (cf. Riesbeck, 1980), especially if (as for most of my freshman) the town is not very familiar.

    2.25 I tentatively classified pauses and hesitations according to their locations in the text (cf. IV.2.15ff). In the following samples, a short pause {164} is marked with “/” a long one with “//”.¹ [In contrast to some studies (e.g, Goldman-Eisler, 1972; Matsuhashi, 1979,1981), I did not undertake to compare the exact length of pauses, given the poorly defined, complicated causalities involved (IV.2.19-21).]  Cross-outs are placed in pointed brackets, and later insertions in curly brackets. Original spelling and punctuation are retained.

   2.26 Writers tend to pause for a choice among alternatives. When the set of options is clearly defined (cf. IVI.9), locations and motives for pauses can be classified at least three ways:

     2.26.1 First, the pause falls immediately before the decision word:  

      (97) Then go straight down Stadium Road until the / Florida Field.

      (98) turn / right into the parking lot

      (99) procede / east toward the town

      (100) there are / < f > three exits for Gainesville

      (101) you reach the intersection of University

      (102) Make a left on 13th / street until you get to // <inner> Inner Circle Road.  

Landmarks (97), directions seen by a person (98) or on the compass (99), the number of exits (100), the names of a street (101), or the designation ‘street’ (vs. ‘avenue’, ‘road’ etc.) (102) present clearly defined sets of options. The writer may not make the final selection until the slot in the sequence actually arrives (cf., II.2.14; III.2.30; IV.1.9; IV.2.10, 15, 27).

    2.26.2 Second, the pause falls before the phrase in which the choice will be made, but not immediately before the decision word:  

(103)  Exit on / the third ramp

(104) Stay on 26 // for about 2 miles

(105)  Stop at Broward Hall //, which is the second group of buildings on the left

 (106)  a shopping center on the right side of the road

(107)  Travel on the road  taking the left fork  

Writers can anticipate and confront choices at some distance from the current point of the text: e.g., a number (103, 104, 105) or a direction (106, 107). 2.26.3 Third, the pause falls both immediately before the decision word and before the phrase that contains it:  

(108) turn left and / go to the / next stoplight

(109) I live / on the / fourth floor  

A phrase involving two choices may have pauses both before and inside it:  

(110) Continue / on Newberry / road  

The examples in IV2.26.2-3 again involve clearly defined sets.

     2.27 Examples (97-110) suggest that discourse choices can be made not only

when the appropriate slot is reached in the sequence, but whenever {165} they are anticipated and confronted. In (103), (104), and (106), the pause comes before the introductory function word (article or preposition), while in (97), (101), and (102), the pause comes after the function word and before a content word; and in (109), we find both cases in a single clause. This variation could arise if the function word itself is a decision point in some cases, but not in others. Or, look-ahead could anticipate decision points in the upcoming phrase or clause and resolve them in advance. The second account postulates a time lag between pauses and decision points: a phrase structure could act as a “distant early warning” system that triggers a set of roughly parallel searches for its constituent elements (cf. IV.2.11). The time expended on these searches would add up to the total pause time. However, if processing is heavily loaded, those searches might be slowed or sidetracked, so that hesitations or pauses would also be needed immediately before the decision point to settle final problems. This account reconciles the dispute over whether words or phrases are selected first (cf. II.2.14; III.2.30; IV.2.10, 15): the order of selection adapts to fit the nature and difficulty of search and decision processes. Thus, the pauses before clauses or sentences partially match syntactic types (IV.2.16), because the latter stipulate the sets of searches to be done; but factors that load the system, such as abstractness, conceptual integration, ambiguities, and reduced coherence, raise the likelihood of pauses irrespective of syntax.¹ [Just & Carpenter (1980) suggest on the basis of eye-movements that in reading, each word is processed as far as possible when it’s encountered. Such a supposition does not fit the hand movements of writing, which are far slower and more cumbersome.] Therefore, the variety of causes listed in IV2.19.1-24 is not contradictory, but natural.

     2.28 The system load would also be increased when decisions confront a vague, poorly defined set of alternatives, because more effort would be needed to determine the goodness of fit between one’s specifications and any available word-choice (cf. III.3.2.5). Pausings would thus be predicted here. For example, visual descriptions of things are more helpful than proper names when a stranger is trying to drive through a town, but are harder to select. My writers had to choose not only the landmark, but also a recognizable description they had probably not had to give before. Pauses could indicate the need to search and evaluate:  

(111) You are at the O’Connel Center, a large white / domed // arena.

(112) turn right just after you pass // the huge white // sports facility, < called > the O’Connell Center

(113) Across the street on Museum road <is> will be two / tall brick towers

(114) There is a road right < before > before < the > Shands /, < you will > which is / noticed by the railroad crossing / sign on the right  

The O’Connell Center that troubled the writers of (111) and (112) is constructed of a special cloth-like substance developed in the space program, and resembles a “Star Wars” station more than a sports arena; {166} no time is needed to tell that it’s ‘large’ and ‘white’, but its bizarre shape makes one wonder what on earth to call it for the guidance of a stranger on campus. The pause in (113) fell not at the decision-word ‘two’, but at the designation for a dormitory. In (114), the writer was having trouble organizing the scene: the landmark ‘Shands’ (Hospital) is not identifiable until after the place where you should turn left; hence the mention of the more visible ‘railroad crossing’ and (after a pause) its conspicuous ‘sign’. (114) illustrates how conceptual indecision, a matter of coherence, can impair cohesion as well.

    2.29 The account suggested in IV.2.27 relates pauses to look-ahead. Pauses before a phrase or clause may signal the act of choosing the phrase format, plus some of the words; pauses inside a phrase or clause may signal problems in searching out and implementing words after the phrase format has been chosen. Nonetheless, some pauses must also be devoted to look-back: inspecting the phrase or clause just composed to see how successful it was and how it will affect what comes next-an instance of open-loop feedback (III.3.2.2). Such pauses would resemble the “rhetorical pauses” identified in speech by Goldman-Eisler (IV,2.18), except that they are for the benefit of the writer, not the audience. Skilled writing might involve many such pauses and thus not appear fluent at all. Careless writing would look faster and smoother, because feedback and evaluation would scarcely be attended to. Therefore, fluency by itself is an inconclusive measure of writing quality (cf. II.2.29; II.3.21, 38; IV.2.34).

    2.30 My filmed writers paused noticeably at the end of most, but by no means all, sentences. Sometimes a writer continued instantly to the next sentence, as if having at least partly planned out the latter while putting its predecessor down on paper. ¹ [Marie Nelson, my colleague who wrote without pauses, reported having deliberately trained herself in college to think ahead in just this fashion.]  At least three ways to omit pauses between sentences seem plausible. First, the slow rate of writing compared to speaking (IV.2.22) could allow on-the-spot planning. When my writers watched their words on the video replay and tried to read them aloud at the same rate they had written them, their speech rhythm kept getting far ahead; they strained to pause before almost every word, just to keep in step. When subsequently asked to read aloud at their normal rate, they paused mostly at the boundaries of sentences or clauses-the usual “rhetorical” motive (IV.2.18, 29). Second, pauses between sentences could be spared if the inscription of the final stretch of a sentence is handled by packages of automatic sub-routines (cf. III.1.24; III.2.31; IV.1.12; V.1.35). These actions wouldn’t interfere with planning out the next sentence, or at least its opening stretch. Whenever the system is not heavily loaded, it could be looking ahead (cf. IV.2.50f). Third, the content of two sentences could be so closely associated via spreading activation (cf. IV.2.47) that planning both at once would be easy, and no pause would be needed between them, as in:  

(115) I worked / closely with / the  AOML on Rickenbacker Causeway in Miami. The job itself entailed editing English language / scientific papers  

On the whole, this writer was not rapid or fluent, witness the choppy sequence. Yet ‘the job itself’ designates content already activated for the first sentence, and no pause was needed. Instead, pauses occur at points where new content was being searched and selected.

    2.31 My film data on writing did not show the correlation between pause length and clause type observed by Goldman-Eisler (cf. IV.2.16). Many pauses within phrases or clauses were as long as or longer than pauses between main clauses or sentences. Even though writing is slower than speaking (IV.2.22, 30), a similar variety of causes appears to influence pausing (IV.2.27). Only if all matters of content, vocabulary, etc. were already resolved could pauses match the complexity of upcoming clauses. My data suggest that writers compose a stretch of text without necessarily having decided the boundaries of clauses or sentences. As they go along, they gather feedback about the current stretch of text and mark off phrase units as seems appropriate. My films revealed pauses between finishing a clause or sentence and placing punctuation after it:

(116) Stop at Broward Hall //,

(117) The papers were the results / of various research conducted / over a / two year period / (1975-76) //;

(118) It <is> believe it is Road 2

(119) Get into the left lane <and> // <until> {when} / you see the V.A. Hospital on the right and Shands //.  

Feedback is used to determine what kind of unit should be terminated. A decision is then made for a comma (116), a semicolon (117), or a period after either a short sequence (118) or a long one (1 9). In (119), the writer paused and changed ‘until’ into ‘when’ before putting the period —  more evidence that the pause served for gathering feedback on the previous text.

      2.32 To decide about clause or sentence boundaries, writers no doubt consider the conceptual integration of text content (compare the “thought units” in Goldman-Eisler 1972: 111, cited in IV.2.18; and the “conceptual content units” in Matsuhashi, 1982: 286f). A well-integrated configuration of concepts maps easily onto a good clause or sentence (cf. V.2.5). However, the processes of conceptual development need not be completed before phrasing can start (III.2.30). Consider the entire sentence whose opening stretch, as we saw in (115), began without a pause:  

(115) The job entailed editing / English language / scientific papers / which various scientists from / this hemisphere // delivered at a symposium in Caracas, Venezuela / in the / winter of 1976.  

Since the passage represents a reasonable “thought unit,” the sentence works out very well. But instead of a long initial pause followed by fluent {168} production of the whole unit, the evidence suggests that concepts (agents, location, time) were still being integrated during the execution of the phrasing. Hence, “thought-units” can be the result of producing the sentence, not just its precondition. The levels of conceptual development, expression, and linearization are organized and chunked in concerted interaction, so that the corresponding production phases run in parallel and influence each other’s progress. This account explains why a long pause followed by the fluent production of a lengthy stretch of text was rare, e.g., in the second sentence of (120):  

(120) Stay on that road / and / go // straight down // until you hit N.W. 13th St. // Make a left on N.W. 13th Street and go up to N.W. 12th Avenue.  

Even my literate academics (with one exception, IV.2.23) fluctuated between fluent vs. hesitant writing of sentences.

     2.33 My students of course had procedural knowledge of how to get from one place in Gainesville to another. But this did not smoothly convert into declarative knowledge (cf. I.4.7). As for many actions (tying a shoe, riding a bicycle, etc.), knowing how to do it was quite different from knowing how to say how to do it. The act of writing made the students take stock of their own knowledge in a new way to make it communicable (cf. I.2.23.5). Instead of going through town themselves, they had to visualize a reader making the trip. Each statement had to be a practical directive whose effect would be to update the scene (cf. I.4.11.2). Some pauses before simple, easily found words, such as ‘and’ and ‘you’, could have been caused by updating:  

(121) Stay on that road / and / go // straight down untill you hit N.W. 13th St.

(122) turn right and // you will / be on S.W. Archer.

(123) Go into the one on the left // and take the lobby <in the> elevator to floor 5, room 508.  

Using discrete language to express the continuity of experience is certainly complex (III.2.25; III.3.19). Perhaps the writer of (123) blurred the event sequence to rush the audience along, as if the elevator could carry the whole lobby to ‘floor 5’ (noticed and fixed) and then roll down the hall to ‘room 508’. The absence of writing pauses in (123) might correspond to this rush of actions — an unusually tight “thought unit.”

    2.34 Even this limited, provisional account of pause types from my film data should suggest the complex correlations between the real time of writing and such mental events as phrasing, memory search, decision, feedback, conceptual integration, updating, and so forth (cf. IV.2.21). As evidence, the pause itself is neutral, often ambivalent. Supportive factors like careful selection and evaluation may create pauses just as much as disruptive factors like anxiety or reduced coherence (II.2.29; IV.2.19.5, 13). And a certain amount of pauses are probably natural side-products of process organization. To judge someone’s literacy according to fluency, though commonly done, is obviously superficial (I.2.8.3; IV2.29). Still, educators {169} should encourage fluent processing by keeping down anxiety, offering approval, tolerating language varieties, encouraging “free writing,” tailoring the instruction to the individual, negotiating with students, and so on (cf. I.2.23.8; VI.1.9; VI.3.12, 27) — conditions easier to meet by using interviews for instruction (I.3.25; III.3.18; VI.3.14; cf. Murray, 1971; Odell, Goswami, & Herrington, 1982). Further research should explore how we can constructively use pauses to monitor and guide text production,

      2.35 The LOOK-BACK PRINCIPLE subsumes all activities in which processing is influenced or controlled by previous activities within the same text or discourse (cf. IV.2.29, 31; IV.3.9, 30ff; V.1.39). As a discourse progresses, current decisions and events are continually affected by those that have already occurred. These constraints, though they may not often narrow down choices to a single possibility, conserve substantial processing resources. If the meaning of a text is a set of hypothesis about appropriate processing actions (I.4.6; II1.2.2; III.2.26; III.3.14), whatever can be done to define that set simplifies the task and reduces cognitive load. Even “the selection of a first word has in greater or lesser degree committed the speaker to a particular construction or at least a set of alternative constructions” (Boomer, 1965: 156) (IV.2.15). The same narrowing of alternatives presumably takes place in all phases of text processing, but not obligatorily (cf. IV.2.41).

    2.36 Involuntary look-back is manifested in reactivation (cf. III.3.33): a word or word-part already implemented appears again. The outcome may be an error in the linearization phases for sounds/letters-both in speech (124) and writing (125) — and for phrasing (126, 127):¹ [Samples are from students except where indicated otherwise.]  

(124) John gave the goy — John gave the boy (Fromkin, 1971: 39)

(125) There we learnead the actual mechanics.

(126) The interactions affect the ability of people to accomplish accomplish their life tasks.

(127) The field goal is done by kicking the ball, which is held up by a team mate, ball through the uprights of the goal posts.  

Reactivations may be immediate, as in (125) and (126), or considerably delayed, as in (127). Either way, they can evidently be overlooked during deliberate proofreading, for at least two reasons. First, the processor might register the presence of the item, but not notice the number of occurrences during production (III.3.33; V.1.45), and then repeat the same processes during reception (cf. III.3.7). Second, the direct contact with the surface text might be so brief that the error doesn’t enter the writer’s mental representation (cf. III.3.7; IV.2.43). The second reason predicts that the shallower the phase, the more likely the error; the first reason does not.

    2.37 Look-back figures prominently in the means for maintaining the cohesion of the surface text (cf. Beaugrande, 1980a: Ch. V; Beaugrande & Dressler, 1981: 54ff; 1.4.12). RECURRENCE is the reusing of surface expressions at a later point, usually an intentional reactivation, e.g.:  

(128) make a right turn on Archer Road. Stay on Archer Road

(129) Make a left on N.W. 13th St […] I go down to the corner, make a left

(130) within a given slice of time different speakers of different ages from different backgrounds speak different varieties of the same language.  

The recurrent expression may refer to the same thing throughout (128) (i.e., may be “co-referent,” cf. IV2.42); or to comparable but separate things (129); or again to a concept shared by various things (130). A related usage is PARTIAL RECURRENCE (Beaugrande & Dressier, 1981: 56f). a part of a word, e.g. the word-stem, is used again, rather than the whole word (“polyptoton,” cf. IV.2.41).  

(131) As a technical editor   I the job entailed editing papers

(132) until getting to the exits for Gainesville. Exit on the third ramp.  

There is typically a shift in word-class, e.g., from noun to verb (132).  Reference can vary, e.g., from agent to action (131), or from a set of objects to an action performed on one of the set (132).

     2.38 Recurrence illustrates one of the characteristic trade-offs in cognition and communication (cf. I.4.15). On the one hand, the recurrent item can be easily processed by merely reactivating roughly the same operations used on it before; effort increases as the items are spaced further apart (cf. Melton, 1970). On the other hand, this ease may impair processing through loss of interest. Apparently, when the system load becomes too light, operations degrade much as they do when load is too heavy: inattention, low motivation, fuzzy operation, and so on. Therefore, the decisive factor cannot be the recurrence itself, but the context of the recurrence. An enriching, dynamic context leads to positive effects, e.g., learning (IV.2.39), emphasis (IV2.40), or rapport (IV.2.41). In this passage:  

(133) Hazel [Gracie Allen’s sister] loved to talk. She was a very outgoing friendly person. She talked to people on the street, she talked to everybody. When she went to the market to buy some tomatoes, by the time she stopped talking to the grocer the tomatoes were out of season. (Burns, 1980: 114)  

the recurrences iconically mimic Hazel’s ‘talking’ and make the story more droll. But recurrences in a static or very slowly developing context have negative effects, e.g. boredom. For example, (134) is distinctly less readable than its revision in (134a):  

(134) You have to learn how to do tricks while barefooting. The only trick I have mastered while barefooting is barefooting on one foot.

(134a) You have to learn how to do tricks while barefooting. The only one I mastered was doing it on one foot.  

{171} Speaking typically tolerates more recurrence because of the greater need to remind oneself and one’s audience of the topic at hand; a student passage like (134) could be a reflection of speech habits (cf. V3.10, 32ff). Student writers should learn where recurrence is or is not to their advantage (cf. Macrorie, 1970: 24ff, on “strong” vs. “weak repetition”; and Van Nostrand, Knoblauch, McGuire, & Pettigrew, 1978: 313ff, on “strategic repetition”).1[Rochester (1973: 84) reports that thought-disordered schizophrenics overuse recurrence to the extreme of making it their only device for cohesion.] Skilled writers find the appropriate ratio between the new and the given, or between the unpredictable and the predictable (IV.2.41).

    2.39 Similar considerations apply to PARAPHRASE, where content remains largely constant while surface expressions are varied (cf. I.4.11.1; II.3.15; VI.1.3) — the redundancy is in concepts, not expressions. The change in expression makes it hard to tell what counts as a true paraphrase. In these passages:  

(135) There was no more any surprise; there was nothing that was not known beforehand. (Govinda, 1976: 206)

(136) I had no idea what they wanted with me, so I just went down there quietly, not knowing what to expect when I got there.  

it is disputable whether the final clause (135) or phrase (136) is a paraphrase of the preceding statement, or else a new statement closely linked to the latter via commonsense causality. In (137), the meaning of ‘drifters’ is rehearsed in two whole sentences:  

(137) Lenny and George’s constant moving about has forced them to become drifters. They have no permanent home and no family. They constantly move from town to town. 

The writer must decide whether the context justifies a possible self-paraphrase. Sometimes, the effect is to retard or impede the development of content (cf. V.3.37; VI.2.31). Tolerant audiences can adjust to conceptual redundancy by ignoring some content (cf. Ehri & Muzio, 1974). Yet paraphrase that enriches or shifts context can support learning better than recurrence (cf. Honeck, 1973; Mandler, 1979).

    2.40 Look-back also figures in PARALLELISM: re-using a surface format with different components. Here, the redundancy is neither in concepts nor expressions, but in phrase-structuring. Processing load is reduced by packaging new content into already activated phrasings, leaving resources free to focus on content and purpose, as in (138), which made stolid George III into a personified natural disaster:  

(138) He has plundered our seas, ravaged our coasts, burnt our towns. (Declaration of Independence) 

Parallelism can be combined with recurrence for special emphasis [though what the US has in 2004 is decidedly NOT ‘government by the people’]:  

(139) government of the people, by the people, and for the people shall not perish from the earth. (Gettysburg Address)  

Conversely, non-parallelism in formatting comparable content can be disorienting. Compare (140) (student’s version) with (140a) (my revision):  

(140) Style to many women is still a mystery, yet to others they have found the secret to the meaning of style.

(140a) Some women still find style a mystery, while others have found the secret to its meaning.  

The contrast between the two kinds of ‘women’ emerges more clearly in (140a) through parallelism (agent + verb + object) and partial recurrence (‘find/found’) (cf. IV.2.51; VI.2.18). ‘Mystery’ and ‘secret could be Fowler’s “elegant variation”.

    2.41 Classical rhetoric developed an elaborate repertory of “schemes” for using recurrence and parallelism (cf. R. Graves, 1974: 187ff). Recurrent items might appear in word-stems (polyptoton, i.e. partial recurrence); in two successive phrases at the start (anaphora), at the end (epistrophe), or at the end of the first and the start of the second (anadiplosis); or, inside a single phrase at the start and the end (epanalepsis). Sound/letter patterns could recur inside words, e.g., at the start (alliteration), at the end (rhyme), or in varied positions (assonance). These schemes were effective in classical oratory partly because audiences expected them and enjoyed the elaboration of the text. But beyond that factor, the schemes presumably affected the cognitive processing of the text itself (cf. Beaugrande, 1979e). The scheme was taken as given, but not its specific use. Hence, discourse events were in part anticipated and seemed fitting and necessary when they occurred; yet they were in part surprising and kept the audience’s attention in focus.¹ [For further treatment of samples of this kind, see also Beaugrande (1978c, 1983b); Beaugrande & Dressler (1981).] Look-back was thus a simultaneous enactment of opposing tendencies — redundancy in some phases (expression, phrasing, sounds/letters) and innovation in others (ideas, concepts). The dynamic tension helped promote rapport between orator and audience. Such effects are general and powerful enough that modern writers might profit greatly from studying classical rhetoric (Corbett, 1971). For example, Christensen (1965: 148) advocated that a series of co-ordinate statements (on the same plane of generality) be parallel, while a series of subordinate statements (descending level of generality) be varied. Resemblances and contrasts in format thereby help to identify the hierarchy of content.

    2.42 While the devices sketched in IV37-41 increase redundancy in the shallow phases, a second set of devices decreases that same redundancy, notably: PRO-FORMS (short, conceptually empty place-holders) and ELLIPSES (incomplete surface structures processed by consulting complete ones in their vicinity) (cf. Beaugrande & Dressler, 1981: 60-69, and references there). {173} Discourse frequently requires CO-REFERENCE: more than one word to denote the same thing in the textual world. The two options just mentioned are much more economical than constant recurrence, and less likely to dull interest. Pro-forms are termed anaphoric if they look back (if the pro-form follows its co-referent),¹ [Co-referent,” though a recent coinage, is precise. “Referent” makes it seem as if the noun referred to the noun, rather than both referring to the same thing in the world; and “antecedent” implies “before,” which need not be the case (IV.2.42).] and cataphoric if they look ahead. Pronouns were the pro-forms most often treated in traditional grammar, but others have been recently discussed (cf. Quirk et al., 1972: 667-700), e.g., pro-verbs (‘be’, ‘do’, etc.) and pro-modifiers (‘so’, ‘such’, etc.).

    2.43 Pro-forms are controlled less by grammatical rules than by the exigencies of linear processing, e.g., how much text can be spanned by look-back signals (cf. IV.2.66; V3.27). Brevity should not be sought at the expense of clarity and comprehension. The preference is for a pro-form to look back to the nearest co-referent available in the surface text (Springston, 1975). Yet even skilled writers override this preference by using proforms for a conspicuous and reasonable co-referent further away. A composition expert wrote:  

(141) Authors who support themselves by writing have the same reluctance about actually getting down to the words-on-paper stage of their job that students do, and overcome their hesitation by using these methods. They set aside a special period of the day [etc.] (S.P. Miller, 1976: 73)  

The ‘they’ in the second sentence looks back to the most distant candidate ‘authors’, not to ‘students’, ‘methods’, or ‘words’. Though ‘methods’ and ‘words’ are plainly unreasonable, ‘students’ is ruled out by the general meaning and purpose of the whole passage. The statement made in the first sentence is to be illustrated in the second, i.e., to depict the ‘Methods used by ‘authors’ (not by ‘students’) to ‘overcome’ their ‘hesitations’.² [The comma after ‘do’ signals that the next verb looks further back to ‘authors’ (not ‘students’) (cf. IV.3.31). The proper co-referent is found by attaching the pro-form to the topic concept ‘author’; if a topic is a control center in the mental representation (III.2.14ff), it would readily attract any doubtful instances of co-reference. However, unskilled writers use vague pro-forms in ways that confuse or distract from the message (cf. III.3.40; V2.25-29).  

(142) By placing a hockey stick in the palm of your hand, it becomes a lethal weapon.  

An entity in the mental representation may engender a pro-form for which the writer entirely forgets to provide an explicit co-referent:  

(143) Physical therapy is a demanding job. They must be both mentally and physically strong.  

{174} The missing co-referent must be inferred by association (the people who take the ‘job’). In such cases, the text producer evidently relies on the retrospective mental representation of prior text and devotes less care to perceiving the surface text (cf. III.3.7; IV2.36).

   2.44 ELLIPSIS cuts down the need for recurrence by simply omitting expressions (Quirk et al., 1972: 538):1 [Grammarians define ellipsis variously, according to what they consider the ideal complete form (cf. Quirk et al., 1972: 538, 574; Clark & Clark, 1977: 16). The definition here is in terms of processing: perceiving an incomplete structure and supplying the rest from a nearby complete structure, or less often, from an inferred complete structure.]  

(144) I’ll pay for the hotel if you will for the food.

(145) You must also be a member of the party, since he is.  

Linear processing may consult the complete phrasing to fill in the missing elements (,pay for’, ‘a member of the party’). Or, the mental representation of prior text may supply the needed concepts directly, without passing through the shallower phases. Ellipsis is kept efficient by certain constraints. A second core (main) clause (146) permits ellipsis more readily than does an adjunct (dependent) clause (146a):  

 (146) Mary was happy and Alice miserable.

(146a) Mary was happy while [because, when, although, etc.] Alice [was] miserable.  

Unskilled student writers who shift phrasing around produce confusing ellipses, e.g.:  

 (147) RCA concentrated on business applications, IBM and UNIVAC on business and scientific and the highest speed computers.  

Like proforms, ellipsis is to save effort, not to waste it piecing things together. Parallelism obviously assists the supplying of ellided items, e.g.:  

(147a) RCA concentrated on business, but IBM and UNIVAC on business, science, and top-speed computers.  

   2.45 The eye doesn’t appear to see very far back from the point where it’s reading (McConkie & Rayner, 1976).² [According to McConkie and Rayner’s experiments, anomalies just three or four letters to the left of the eye fixation point seem to go unperceived.] Look-back could be done by simply scanning the linear sequence of the surface text backwards, but that seems unlikely. Eye-movement studies indicate that readers fixate the specific prior words that are most likely to assist current processing (cf. Just & Carpenter, 1980). For example, to find the co-referent for a pronoun, the eye selectively focuses on the most plausible preceding nouns. A mechanical backward scan would waste effort on many irrelevant words. Moreover , as I noted in IV.2.43, deeper phases than phrasing are consulted,  especially when the co-referent for the pronoun is remote, vague, or missing. {175} Perhaps the processor does a single all-purpose look-back operation to meet current requirements, like a bus carrying passengers watching for their respective stops (compare Fahlman’s 1979 memory model). The processor may routinely consult the mental representation of prior text and re-scan the surface text only on strategic occasions, e.g. for revision (IV.2.49f). The eye movements of readers and hand movements of writers go backward relatively seldom in comparison to the extensive need for look-back to maintain cohesion and coherence (cf. Matsuhashi, 1982: 279). Besides, the spoken modality scarcely permits any such backward surface scans, whence its higher ratio of redundancies and inconsistencies (V.3.10). Consulting the mental representation, though conserving on the surface, may have negative side-effects such as pre-activation and re-activation (IV.2.36, 47), where the processor appears to lose track of the surface text momentarily. The same cause could account for an audience’s tendency to confuse what was said with what they supplied during comprehension (cf. III.1.14; III.3.7; III.3.27).

    2.46 Such tendencies (along with those treated in IV.2.47ff) yield clues about the relation of the linear surface text to overall processing. The mental representation of the text, being a memory store, necessarily shares the general properties of memory itself. Hence, the representation would adapt to the perspectives adopted by the processor in each context (cf. III.3.10). This effect would fit the nature and extent of the processes themselves, e.g. their motivation, duration, thoroughness, and so on. At various times, the representation could assume an order of dimensions such as (a) linear (to reflect the surface sequence); (b) hierarchical (to reflect degrees of generality and importance); or (c) categorical (to reflect the classifications or associations obtaining among concepts, expressions, phrase formats, or sounds/letters). When a given dimension becomes relevant, the corresponding perspective becomes active. Admittedly, such a versatile mental representation is much harder to formalize than sentence structures or logical forms (cf. J. Anderson, 1976, vs. Beaugrande, 1980a, 1982e), but it appears justified in view of recent findings on processing and memory.

     2.47 The LOOK-AHEAD PRINCIPLE is the complement and converse of look-back, and subsumes all activities directed to subsequent parts of a discourse. Aside from deep-level goals, look-ahead ought to have lower priority than look-back, because the latter can compensate for the former more readily than vice-versa: it’s easier to make the current decision fit previous recorded ones, than subsequent anticipated ones. Hence, look-ahead would tend to be sacrificed under heavy system load (III.3.5.5; IV.2.50f). Still, the operations of look-ahead ought to resemble those of look-back. For example, look-ahead should also operate on a versatile, multi-phase mental representation (IV.2.45f) — the more so as the surface text per se is not normally available in advance. Some sounds, words, phrases etc. may be correctly anticipated, but not a complete, detailed text {176} (III.3.5.5; III.3.7). How rich, specific, flexible, or conscious such expectations may be is far from established. At least some look-ahead is automatic, thanks to spreading activation (IV.2.48), but is likely to make more material active than is relevant to the ongoing discourse (III.2.20f). Both the producer and the receiver of a text must have some screening process that scans memory associations for relevance (cf. IV.2.49; VI.2.32). If look-ahead is especially sensitive to heavy processing loads, screening could easily malfunction. For instance, just as involuntary look-back is documented in reactivation (IV2.36), involuntary look-ahead is documented in pre-activation. Linear errors can appear on the level of sounds/letters, both in speech (148) and in writing (149), or on the level of phrasing (150, 151):  

(148) the sird and-the bird-the third and surviving brother (Fromkin, 1971: 30)

(149) All basketballs game are started on jump shots

(150) The Federal Defense Budget is the only government that program will be increased.

(151) take the lobby < in the > elevator

Unless corrected (148, 151), the error may seriously impair cohesion (150) and coherence (151). On a larger scale, pre-activation may start a whole clause too soon:¹ [On the bracketing conventions, see IV.2.25.]  

(152) go east <until coming/to> //(you are now on {West l University Blvd.) // until you come to 20th St.

(153) Turn left / <stop at the stop sign,> {follow / the / road) and turn /1 right into the parking lot / of the hospital. // <you > The lot is // immediately after the stop sign.  

In (153), ‘stop at the stop sign’ was not deleted until after ‘turn right’ was inscribed; there was a pause before ‘right’ and a slowdown on the word itself. During another long pause after ‘hospital’, the hand moved back to cross out ‘stop sign’ with more ink and to insert ‘follow / the / road’; only then was the comma after ‘hospital’ converted to a period and a new sentence was begun. These linear rearrangings, unlike those in (148-151), are conscious, voluntary acts of look-ahead.

    2.48 The impact of look-ahead is also manifested in MISCUES, where people alter a text as they read it aloud (cf. Goodman & Burke, 1973). The same rapid, largely automatic processes that constantly engender hypotheses about what is going on in the world naturally make occasional mistakes (Neisser, 1982; cf. 1.1.7). Readers sample the printed text while at the same time guessing what words are probably on the page (Goodman, 1970; Frith, 1980). The most immanent miscues are those where the interpolated word both looks like the correct one and makes sense in context {177} (cf. III.3.37). Students who had already closely studied the texts containing (154) and (155) nonetheless miscued when reading aloud to a class:  

(154) Through dust where the blacksnake dries [dies] (James Dickey)

(155) My crop of corn is but a field of tares.

         And all my good is but vain hope of grain [gain]

        (Chidiock Tichborne)  

Associations such as ‘dust-dry’ and ‘corn-grain’ easily come to mind via spreading activation in people’s memory for world-knowledge (cf. I.4.11.2; III.2.14, 20). Miscuing even on familiar texts again shows how performance need not reflect real knowledge or ability (cf. I.3.3). Miscues are the sign of efficient readers; children who slavishly cling to the printed page generally do poorly on comprehension and learning of content (Goodman & Goodman, 1979). When I recorded college students reading aloud, many were unable to recall the content of simple passages because they were “concentrating on the words” (their explanation). If expressly told to read for content, they regularly produced miscues, especially ones I had designed into the samples. One version of my ‘rocket’-text (Beaugrande, 1980a: 222ff) contained contrived oddities:  

(156) With a roar and a burst of flares, the giant rocked on its pad.

(157) It plunged down to the earth’s surface 40 miles from the launching padded by landing gear.  

Almost all readers miscued on (156) with ‘giant rocket’ (one person three times in a row) and on (157) with ‘launching pad’. Even persons who did not miscue later reported having read ‘giant rocket’ and ‘launching pad’. ‘Rocket’ was the topic concept high in the hierarchy of the mental representation and thereby dominating look-ahead and look-back. Readers thus miscued although there were 84 words between ‘rocket’ and ‘rocked’. The ‘roar’ and ‘burst’ were at once construed as results of the rocket’s take-off.

    2.49 Look-ahead would operate in a special way when written texts are revised. The surface text is now available for consultation alongside the predictive mental representation of it. Current decisions can be better aligned with anticipated ones. Processing load is generally lower than during the original act of production, because many tasks are provisionally done. In terms of the scheme in III. 3.2, there is a rise in familiarity and perceptual immediacy (visible text), and a lower demand for motor control. Feedback is easier to obtain, and attention can be focused on a few specific problems, rather than diluted over a great many. Lower processing load in turn creates more favorable conditions for look-ahead (cf. IV.2.47). The writer can inspect results and detect noteworthy departures from the communicative intention. Larger chunks can be formed in order to compare a specific point with its general context. Content can be judged for relevance and interest, so that trivial or pointless associations that had automatically come to mind can be consciously screened out (IV.2.47). Preference should {178} go to variable or problematic content that elicits interest (cf. III.2.24; IV.2.11; VI.2.2, 8f, 32). Such content might carry special markings in memory storage to alert the scanning techniques (cf. Schank, 1977).

    2.50 A sub-type of revision can be called RESTARTING: a stretch of text is not finished, but started over (cf. IV.2.15; V.2.34; V.3.20f). Processing presumably reactivates operations, adding conscious effort to resolve problems or blocks. My film showed writers altering phrase-initial function words, either right away (158), (159), or after the phrase was done (160):  

(158) There is a road right < before > before < the > Shands

(159) Once there, turn right / < until > on university until you come to the intersection <on> at Museum road.

(160) Get into the right lane <and> < {untill} (when) you see the V.A. Hospital

The initial word was used after all in (158), and saved for later in (159). Under proper conditions, look-ahead can apparently judge the appropriateness of phrase-initial words quite soon. This faculty operates less well over a larger stretch of text. In (161), the writer crossed out ‘w’ for ‘we’ at once, opting for an active over a passive construction to put the ‘horrifying’ cause in focus (cf. II.3.12.2; IV2.61). Yet after writing ‘frightens us’, he made a long pause, went back, and changed ‘horrifies us’ to ‘we enjoy being horrified by’ — the experiencers were judged more in need of focus than the causes of the experience. The original version was inconsistent, because the crucial distinction (being ‘horrified’ vs. ‘enjoying’ it) was not made clear soon enough.  

(161) I am interested in what <w> <horrifies us> {we enjoy being horrified by}, in what we communally and voluntarily seek out to be thrilled by: I am not interested in what really frightens us.  

Whether revision can profit from restarting depends on the writer’s skill. Unskilled writers are prone to work without very much look-ahead, the more so if the latter degrades under heavy processing load (cf. IV.2.47, 51). These writers tend to restart so often that they get writer’s block long before a paper is finished (cf. III.2.9, 17, 20; III.3.4). If they could hold off until they have a fairly long stretch of text, at least a paragraph, they would draw more profit from restarting. At that time, context and purpose are better defined to provide direction and criteria for decision-making. Skilled writers are better equipped to look ahead, and hence to restart brief passages.

    2.51 The cohesive devices noted for look-back in IV.2.37-44 also apply to look-ahead. Recurrence, paraphrase, and parallelism may be planned such that the prior passage anticipates subsequent ones. Shakespeare’s sonnet XX beginning:  

  (162) A woman’s face, with nature’s own hand painted,

Hast thou, the master mistress of my passion;

A woman’s gentle heart, but not acquainted

With shifting change, as is false woman’s fashion;  

uses artful recurrences of ‘woman’, emphasized with parallelism in line openings, to look ahead toward a surprising contrast when the addressee turns out to be a man (cf. IV2.40; VI.2.18). The streamlining devices of proforms and ellipsis can look ahead cataphorically (IV.2.42). The proform then precedes the expression providing the content:  

(163) Because they go into the blood directly instead of through ducts, hormones circulate all over the body.

 The reverse (anaphoric) placement would be ambiguous:  

 (163a) Because hormones go into the blood directly instead of through ducts, they circulate all over the body.  

Ellipis can also be cataphoric:  

  (164) Because Alice won’t, Mary is dusting the furniture. (Quirk et al., 1972: 538).  

Cataphoric uses of pro-forms and ellipsis are much rarer than anaphoric. Look-ahead seems to be more limited than look-back, presumably because the encounter with the surface text has not yet occurred (cf. III.3.5.5); and because look-ahead would be likely to suffer immediately when processing is heavily loaded (cf. IV2.47, 50).

    2.52 JUNCTIVES are a word-class that links a forthcoming stretch of text to the currently ending one. Junctives contribute to look-ahead by indicating the nature of that linkage. CONJUNCTION (e.g. ‘and’) is a neutral signal of addition. DISJUNCTION (e.g. ‘or’) announces alternatives. CONTRAJUNCTION (e.g. ‘but’) prefigures unexpected or contrastive content. SUBORDINATION signals a range of dependencies, notably, causality (e.g. ‘because’) or time proximity (e.g. ‘whenever’). In strict classifications, subordination prevents a clause from having sentence status (cf. V.2.24, 36, 40). If a subordinating junctive opens a sentence, an independent clause is expected later on. The other three types of junction are not so constrained, and can appear freely at the start of sentences as well as inside them (V2.26).1 [Unless finicky “rules” are passed against junctives at the start of a sentence (I.2.14).] Disjunction offers dual signals (either/or’) to alert the audience right away that alternatives will be presented.

     2.53 The HEAVINESS PRINCIPLE concerns the extent to which materials in discourse impose a load upon processing, because they are important, salient, complex, unfamiliar, unexpected, and so on. Though logically diverse, these aspects have much the same operational effect of loading the {180} system. The heaviness principle can interact with the core-and-adjunct principle, but the two are not identical. Any ongoing structure has central vs. peripheral components (cores and adjuncts), but only some components or whole structures create a heavy load on processing in a given context. A core need not be heavy, nor must every heavy element be made a core; these correspondences are strategic —drawing control toward items that are important for two reasons — but not necessary. For example, the subject of a sentence is a core element, but often conveys familiar content (IV.2.63f). And, some adjuncts are made heavy by getting moved outside the structure containing their cores (cf. IV.2.58; V.2.32; V.3.30).

      2.54 In terms of information theory, heaviness depends on the degree of UNCERTAINTY (cf. IV.1.1f). Every act of communication removes some uncertainty about what will be said. Statistics tell us little about the experience involved (IV.1.2), particularly since the surface text is a sparse document of only those alternatives that were finally chosen, not those that might have been. Heaviness results from a fairly improbable choice within a rich context, as in this narrative by Franz Kafka (in Landy [Ed.], 1980: 246f):  

(165) I must have coal; I cannot freeze to death; behind me is the pitiless stove, before me is the pitiless sky, so I must ride out between them and on my journey seek aid from the coal dealer. But he has grown deaf to ordinary appeals; I must prove to him irrefutably that I have not a single grain of coal left. [... ] My mode of arrival must decide the matter; so I ride off on the bucket. Seated on the bucket, my hands on the handle, the simplest kind of bridle, I propel myself with difficulty down the stairs; but once down below my bucket ascends, superbly Through the hard frozen streets we go at a regular canter [etc.]  

Readers don’t expect the narrator to ‘ride off’ on his coal-bucket; but in context (need to prove the bucket is empty; inability to afford a real horse), the action seems motivated, and is soon integrated into the story line, making later events seem natural (‘cantering’ through the streets, etc.).

     2.55 NEGATION is a useful gauge of uncertainty, because it usually repudiates what would otherwise be assumed (cf. Wason, 1965; Osgood, 1971; Beaugrande, 1978c; Givon, 1978). (161) rejects the notion that ‘horror’  includes ‘what really frightens us’; (162) makes an exception to its own overstatement that a ‘woman’s heart’ is ‘acquainted with shifting change’; (164) would be appropriate only if there were reason to expect ‘Alice’ to ‘dust the furniture’; and’so on. Languages typically associate positives with good, and negatives with bad (Clark & Clark, 1977: 538).¹ [Boucher & Osgood (1969) found that many languages add negating affixes to good evaluation words, but not to bad: ‘unhappy’ vs. ‘unsad’, ‘unclear’ vs. ‘unmuddled’, etc. This favoring of the good gave rise to the “Pollyanna” hypothesis, named after the relentlessly cheerful heroine of Eleanor Porter’s 1913 novel of the same name.] By an odd coincidence, composition instruction has a traditional bias {181} against multiple negation, either as “wordy” (166) (cf. Williams, 1981a: 158) or as a “stigmatized” (167) usage, e.g., in Black English, which, similar to such languages as Russian, has “negative concord” (Labov, 1970) (cf. Shaughnessy, 1977: 93):  

(166) Housing for married students is not unworthy of consideration.

(167) It won’t make no difference.  

In everyday language, multiple negatives intensify, rather than cancel, each other

(I.2.16.8). Multiple negatives are hard to process only if they do cancel each other, because having one’s expectations repeatedly violated increases heaviness (cf. Carpenter & Just, 1975; Sherman, 1976). Each negation removes us again from our point of orientation. The Duchess’s injunction (Carroll, 1960: 122)  

(168) Never imagine yourself not to be otherwise than what it might appear to others that what you were or might have been was not otherwise than what you would have appeared to them to be otherwise.  

 is impenetrable (not just to Alice) with five negations (plus a recurrent ‘otherwise’ at the end that is not a negation). When four are removed, the content may come into reach:  

(168a) Never imagine yourself to be what it might appear to others that you were or might have been under other conditions.  

that is, ‘always see yourself as you are under real conditions’.

    2.56 SALIENCE can be defined as the quality of impinging strongly on the senses. Among all the things available in perceptual space (the totality of what is perceived at any moment, III.32.10), salient ones draw focus: extremes in size, color, brightness, loudness, mobility, novelty, and so on. Osgood (1971; cf. Osgood & Bock, 1977) undertook to prove that salient elements tend to be expressed earlier in an utterance or sentence than nonsalient ones. For example, mobile objects would be mentioned before stationary ones — an illustration of the “normal ordering strategies” enumerated in III.3.19. Osgood placed a cardboard tube upright on a plate, and 21 out of 26 test persons said things like (169) rather than (169a); 23 out of 26 mentioned a mobile ball before a stationary one despite the latter’s large size and bright color, e.g. (170) rather than (170a) (Osgood & Bock, 1977: 107f):  

(169) A cardboard tube is on the plate.

(169a) The plate has a cardboard tube standing in its center.

(170) The small black ball hits a very big orange ball.

(170a) A very big orange ball is hit by a small black ball.  

Evidently, the different aspects of salience can compete with each other in determining linear formatting (cf. IV2.62).

   2.57 IMPORTANCE is another gradation of heaviness. Materials may be important for the text producer’s goals, but not for the receiver’s. Goals may interlock (e.g., entertain vs. be entertained), conflict (e.g., deceive vs. not be deceived), or simply be unrelated (e.g., deliver a Hyde Park speech to no audience vs. visit the Tower of London). Each of these situations affects the organization of the text, and the heaviness of its elements. In this text from a stage play (cf. Beaugrande, 1980a: 183ff), the speaker imposes her goal on a co-operative group:  

(171) We got one thing, and one thing only, to do now. And that is to find out if Abby’s planning to take that portrait to Chicago with her.  

The important goal of getting Abby’s ‘portrait’ calls for creating focus with dummy (place-holder) expressions (‘one thing’, ‘that’) which look ahead to heavy elements (IV2.63), and then announcing a main plan step (‘finding out’ etc.) Notice that postponement can signal heaviness just as well as early mention.

     2.58 INVOLVEMENT can designate the intensity with which people are participating in discourse interaction. In the studies cited in IV.9.56, Osgood argues that involvement also moves elements earlier in the sequence. Laboratory experiments — e.g., describing to an unseen child what Osgood did with a shopping bag full of items on a table — seldom elicit much real involvement. An everyday illustration is found in this science fiction story, where the absent-minded pilot is threatened with disaster unless he finds the ‘cribsheet’ with the flight code he was supposed to have memorized:  

(172) He suddenly went numb all over. “The crib! Where’s the cribsheet! “ That was the awfully important detail. (Lem, 198 1; 16)  

 His shout begins with the item of greatest involvement moved outside the sentence structure in which the question word would come first; the complete format is then uttered.¹ [On the tendency to create sentence fragments for emphasis, see V.2.32.] The heaviness principle has moved an element outside of its normal core-and-adjunct pattern (cf. IV.2.53; V.2.32; V.3.30). Ertel (1977) suggested that “ego involvement” can decide which objects in one’s perceptual space get expressed in subject slots of sentences. The object nearest the “ego” becomes the point of orientation for planning the sentence. In 26 out of 32 newspaper reports about soccer matches, he found a consistent tendency to mention players of one’s home team (close to one’s “ego”) in subject position and players of the opposing team in the predicate of the sentence (1977: 156f).

   2.59 Uncertainty, importance, involvement, and salience naturally interact. Someone beset by grave uncertainty can become heavily involved, {183} and the resolution can be experienced as a salient event. Many text producers choose such devices as suspense, shock, and surprise to heighten audience involvement, especially if the text’s main purpose is entertainment. It would be hard to distinguish, as Osgood and Bock (1977: 90) want to do, between the inherent heaviness of discourse materials (e.g., objects in a perceptual space) and the heaviness assigned by a processor. ¹ [If conflicts arise between the environment and the human, Osgood wants to show that the former wins out, since he favors a perception-based organization for all language (II.2.23). But we need evidence from more natural settings.]  What counts as salient in an object depends in part upon how humans conceptualize the normal values for the object. A bright orange color and a size of two feet across are far more salient for a golf ball than for a beach ball.

    2.60 In speech, heaviness can be distinctly signalled with intonation, pitch, stress, and volume. A heavy element like ‘crib’ in (172) would be uttered in a loud, emphatic voice, accompanied by salient gestures and facial expressions. David Brazil (1975) presented a system for correlating the heaviness of discourse elements with a rise or fall in tone. A passage presenting new or uncertain knowledge has a falling intonation curve; strong involvement would show first a rise, then a fall (cf. also Halliday, 1967; Coulthard, Sinclair, & Brazil, 1978). Writers can use special print (italics, bold, block capitals, etc.) and punctuation (question marks, exclamation marks) to reflect heaviness, notably in transcribed speech (IV.3.18ff). But these devices soon lose their effect. A subtler method is to format heavy materials in short, simple clauses and sentences, as in (173), the finish of a dramatic science-fiction yarn (Lem, 1981: 37). For added effect, each sentence forms its own paragraph (cf. IV.3.17; VI.2.15). The second row of spaced periods adds a split second of suspense about the identity of the pilot (cf. IV3.25).  

(173) Then, from out of the cabin staggered a writhing hulk of a man in a brown uniform, his helmetless head bobbing around like a blurry blotch, his face contorted in a mute shriek ...

Pirx’s knees buckled.

It was ... Boerst.

He had crashed into the Moon.  

Familiar materials stated in short sentences seem more important than otherwise, e.g., in this text beginning (cf. Beaugrande, 1977b: 329):  

(174) Light is radiation that we can see. Heat is radiation that we can feel. Now we are hearing more and more about another kind of radiation as a result of man’s continuing scientific and engineering achievement. This is nuclear radiation.  

This text was distributed by the Atomic Energy Commission to mislead the public that nuclear energy is safe and reliable. Recurrence and parallelism {184} enhance familiarity and make the message appear fitting and inevitable (cf. IV.2.38) by the time ‘nuclear radiation’ is mentioned (again in a one-sentence paragraph for greater effect). Shortness of sentences goes well with these intentions.

     2.61 Heaviness interacts with phrasing to distinguish between active vs. passive sentences. Grammar alone might predict that passives, being longer and more complex, are harder to produce and understand than actives (cf. Coleman, 1964). Composition teachers and handbooks traditionally disparage or even forbid the passive (I.2.14; VI.I.22). Yet cognitive processing favors this format when the subject of the passive sentence outranks the agent (the subject of the corresponding active) in (a) animateness (H. Clark, 1965); (b) size (Johnson-Laird, 1968; Flores d’Arcais, 1974); (c) earlier perception (Prentice, 1967; Turner & Rommetveit, 1967); (d) focus of attention (Tannenbaum & Williams, 1968; Olson & Filby, 1972); (e) imageability (James, Thompson, & Baldwin, 1973); or (f) prior mention (Hupet & Le Boudec, 1975). These findings caution against a narrowly grammatical evaluation of linear patterns that should rightfully be included in the repertory of learners (cf. Walpole, 1979).

   2.62 The linear ordering inside written clauses and sentences is prominently determined in English¹ [On some other languages, see the literature cited in IV.2.63.] by an interaction between the heaviness principle and the core-and-adjunct principle (cf. Beaugrande, 1977a, 1979b). Cooper and Ross (1975) collected standing phrases in which the element closer to the “ego” (“me”) precedes a more remote one (cf. IV.2.58). This ego is stereotyped as “here, now, adult, male, positive, singular, living, friendly, solid, agentive, powerful, at home, and patriotic” — a regular “Archie Bunker” (1975: 67). Hence, we find ‘this and that’, not ‘that and this’, or ‘come and go’, not ‘go and come’ (“here” dimension); ‘sooner or later’, not ‘later or sooner’ (“now” dimension); ‘husband and wife’, not ‘wife and husband’ (“male” dimension); and so forth. Fillmore (1977) sorts out much-debated sentence pairs like:  

(175) I smeared mud on the wall.

(175a) I smeared the wall with mud.  

where only (175a) implies the entire wall was covered. The difference is due to placing ‘wall’ in the core-unit slot of direct object in (175a), as opposed to the adjunct slot of phrasal modifier in (175); the core is readily taken as the focus of concern. A related finding emerged from Linde and Labov’s (1975) study of how people describe their apartments from mental maps. By and large, major rooms (usual points of orientation) were mentioned in core slots (e.g. subject or direct object), while minor ones were mentioned as adjuncts (e.g. prepositional phrases) as in (1975: 937):  

(176) I entered a large living room with the kitchen off the back.  

In Pike’s description of his boyhood home (III.3.25), the house is the topic, and figures in the subject slot of five of the ten independent clauses. The ‘barn’, ‘woodshed’, ‘homes of the preacher and store keeper’, ‘general store’, ‘schoolhouse’, and ‘church’ are all mentioned in predicate slots. Cooper and Ross’s “here” dimension seems to influence whole passages, not just standing phrases.

    2.63 In linguistics, the impact of heaviness on the linearization of sentences has been designated “functional sentence perspective,” “theme vs. rheme,” or “topic vs. comment” (cf. Halliday, 1967-68; Chafe, 1970; Sgall et al., 1973; Danes [Ed.], 1974; Grossman, San, & Vance [Eds.], 1975; Li [Ed.], 1976; surveys in Chafe, 1976; Firbas & Golková, 1976; Jones, 1977: 52-101; Beaugrande, 1992). In many languages, the early stretch of a sentence format, often the subject, is routinely used to express the point of orientation (known or predictable content) whereas heavy materials are mentioned late in the sentence, often in the end stretch of the predicate. A beginning with heavy materials is reserved for special emphasis (salience, importance, involvement, IV.2.56-58). Thus, heaviness normally follows look-ahead, but in emphatic cases, follows look-back. The so-called “cleft” construction¹ [The term “cleft” construction goes back to Otto Jespersen (1924:147f): “A cleaving of a sentence by means of ‘it is’ serves to single out one particular element of the sentence and very often, by directing attention to it and bringing it, as it were, into focus, to mark a contrast.” The “cleaving” is seen by comparing (177) to (177a) and (177b). The “pseudo-cleft,” e.g. (178), doesn’t actually “cleave.”] starts with a dummy subject and predicate ‘it + be’ to delay communication momentarily (cf. IV.2.57) and to reserve the high-focus predicate nominative core slot for items that would otherwise appear as adjuncts and attract less attention. Contrast (177) against (177a) or (177b) (Quirk et al., 1972: 951):  

(177) John wore his best suit to the dance last night.

(177a) It was last night that John wore his best suit to the dance.

(177b) It was to the dance that John wore his best suit last night.  

In return, the materials in the relative ‘that’-clause are taken as obvious, known, or non-heavy (II.3.41). The “pseudo-cleft” construction does the reverse: putting lighter materials before heavier ones. Compare:  

(178) We want Watney’s

(178a) What we want is Watney’s  

The first version of the slogan, though shorter and denser, proved far less durable and effective (Halliday, 1967: 224). The second one focuses on the ale brand by announcing that something is ‘wanted’: a momentary vacuum is created into which the brand name is then placed. (And when you’ve had a few, you can get a hearty assonance in some dialects among ‘what - want -Wat’)

     2.64 As we see, heavy materials can migrate either toward the end of a sentence or toward the beginning. If these tendencies conflict, the outcome {186} is decided by participant roles, according to Osgood and Bock (1977: 93). If the text producer is the center of concern, heavy elements come earlier; if the audience is the center, heavy elements come later. Early mention reflects the producer’s immediate reactions; late mention is a better guideline for the audience.¹ [Unskilled writers often use non-strategic sentence formats that fail to guide focus. This tendency may be related to their strong ego-involvement vs. that of the audience (cf. Lunsford, 1977, 1981; Odell, 1977; Maimon 1979).] Or, the producer may elect to place the point of orientation in the sentence subject, irrespective of heaviness. Consider:  

(179) Two harlots came to the king and stood before him. The one woman said, “Oh, my lord, this woman and I dwell in the same house; and I gave birth to a child while she was in the house. Then on the third day after I was delivered, this woman also gave birth; and we were alone; there was no one else with us in the house, only we two were in the house.” (1 Kings 3: 16-27)  

The subject of the first sentence conveys new content, ‘Two harlots’, whereas ‘the king’ Solomon is already familiar from previous stories in the book of Kings. But subject placement helps to announce the topic characters for this particular story (point of orientation), whereas the predicate gets them onto the scene. The same two characters, taken together or singly, fill the subject slots of the next four clauses: ‘one woman’, ‘this woman and I’, ‘I’, and ‘she’. New, important events and situations are announced in predicates of main clauses: ‘came to the king’, ‘dwell in the same house’, ‘gave birth to a child’, and ‘were alone’. Predicates whose content is predictable appear in dependent clauses: ‘while she was in the house’, and ‘after I was delivered’. The next two assertions use paraphrase to rehearse content which, though already known, is crucial for legal proceedings when the two sole witnesses make contradictory reports. This contradiction creates the suspenseful problem the king will resolve with his ingenious judgement. Thus, the phrasings are subtly designed to look back and ahead across various spans, regulating the ratio of the known vs. the new, of stability vs. innovation (cf. I.4.9).

     2.65 The DISAMBIGUATION principle subsumes all activities of reducing or precluding multiple conceptual configurations for a single stretch of surface text. AMBIGUITY can designate an unintended multiplicity of meaning, and POLYVALENCE an intended one (Beaugrande & Dressler, 1981: 84). Only the former counts as an impediment to communication, whereas the latter is an enrichment, especially in literature (cf. Empson, 1930; Kris, 1952; Dillon, 1978). Disambiguation was extensively treated in mentalist linguistics because of its central role in philosophical semantics. It is no coincidence that Katz and Fodor (1963: 174f) started their frail “semantic theory” with the ambiguous utterance:  

(180) The bill is large.  

They argue that “semantic theory” must “reconstruct” people’s ability to “detect non-syntactic ambiguities and characterize the content of each reading of a sentence.” Yet they reject the operational solution of studying how the context determines the meaning because — true to the philosophical, classifying methods of linguistics (I.4.3, 13; II.1.5) — they think such a theory would need to “represent all the knowledge speakers have of the world” (1963: 178). Instead, they prefer the mentalist approach of assigning disambiguation to “deeper” levels of structural analysis (cf. II.3.6f).¹ [Occasionally, transformations can be harnessed to make grammatical ambiguities clear without technical explanation. For instance, whether an ‘-ing’ form is a predicate modifier or a part of a present continuous (or “progressive”) verb can be made clear by converting to ‘should (not)’ + infinitive. ‘The drunks were staggering’, for example, can be converted to The drunks should not stagger’; hence the participle belongs to the verb. ‘The bill was staggering’ cannot be so converted; hence the participle is a modifier.] In the procedural outlook, the quantities of ambiguities found by such analysis far exceed those that are relevant in real discourse. The question is then: what ambiguities are processed, and how are they precluded or resolved?

     2.66 Processors must have skillful operations that relate any element or linear sequence to context and communicative setting. Hence, the disambiguation principle is fully integrated with the other principles of linear processing — indeed, largely a by-product of their functions. Look-back and look-ahead determine how a current stretch of text should be interpreted in view of previous and subsequent ones. Heaviness marks off the point of orientation and the important material. Ambiguities are thus constrained by expectations arising both from the text itself and from shared cultural knowledge. For instance, within the Pike text (III.3.25), the ‘it’ in (53.2, 3, 4) obviously looks back to topic concept of ‘house’ (IV.2.62), though the surface text has competing nouns nearby (‘barn’, ‘stove’, ‘mansion’, etc.). Commonsense reasoning about goals obviates grammatical ambiguities in a surface text like this (after Wilensky, 1980: 28):  

 (181) John wanted Bill’s bicycle. He walked over to him and asked him if he would give it to him. He refused. Then he told him he would give him five dollars for it, but he would not agree. Then he told him he would break his arm if he didn’t hand it over. He didn’t want to get hurt, so he gave him the bicycle after all. 

Though careless, the text is not at all incomprehensible. A reader imagines an everyday bargaining situation with two character roles pursuing their respective goals, and easily attaches each pronoun to one or the other. No reader would stop at each pronoun to compute all logically possible combinations for ‘Bill’ or ‘John’. Readers want to maintain coherence, not postpone it.

     2.67 Still, ambiguity is a real communicative obstacle on certain occasions. If a topic or context is withheld, comprehension and memory suffer {188} drastic setbacks (Bransford & Johnson, 1973). Ambiguity may be consciously experienced and resolved, especially if it is persistent and salient. Or, it may remain unconscious, barely on the peripheries of working memory.¹ [Presumably, some portions or features of the items must get into working memory, enough to increase processing load, but not enough to reach the threshold where they would be consciously registered (cf. III.2.4; III.3.7; Marcel, 1980; Morton, 1980).] For example, the several senses of a word often go unnoticed in context, but behavioral signals such as hesitations, restarts, or errors indicate they may all be activated (cf. MacKay, 1966; Conrad, 1974; Warren & Warren, 1976). This activation, though automatic, nonetheless seems to increase processing load. Communication could register and resolve ambiguities on many levels of processing and degrees of conscious awareness. A careless text producer leaves this task to the audience — biasing the trade-off in effort noted in III.3.16. Different frameworks of reference may also be the culprit. Soon after I arrived at the University of Florida, I asked my secretary, ‘Where are the alligators? (the campus did have some live ones on the grounds). Her reply, ‘We keep them on the coffee table in the lounge’ created an alarming mental image until I learned that she meant the school newspaper (The Florida Independent Alligator).

      2.68 Such incidents show that whether an ambiguity is present is less crucial than whether it comes into focus. The “Lower Case” page in the Columbia Journalism Review displays unnoticed ambiguities from hastily written newspapers and thereby makes them salient:  

(182) Teachers Strike Annoying Students (Palm Beach Post, Nov. 21, 1979)

(183) Some fossils said to back creationism (Chicago Sun Times, Dec. 16, 1981)

(184) Police Officers Finally Getting Shot at Promotion Exams (Atlanta Constitution, May 4, 1981)

(185) Cooper feels secretaries more than clerks (Fort Collins Coloradoan, Feb. 24, 1980)

(186) Utah Girl Does Well in Dog Shows (Salt Lake Tribune, Dec. 30, 1981)

(187) Reagan wins on budget, but more lies ahead (The Morning News, Blackfoot, Idaho, May 22, 1981)  

As we see, the point is not whether readers can puzzle out the sense, but whether the writer can afford to be careless and inconsiderate (III.3.17; V.3.29). A student’s passage like:  

 (188) I hated eggs, but these people were so hospitable that I had to eat them.  

 is comprehensible, but may leaves the impression of an unskilled or negligent writer (III.3.40). On the other hand, my students sometimes create seeming ambiguities expressly to increase interest, e.g., this description of Mardi Gras:  

(189) One of the biggest things to do In New Orleans is sucking heads and pinching tails. I am sure you guessed that I am referring to eating crawfish.  

A logical deficiency can be an operational advantage if strategically applied (cf. IV.2.55).

    2.69 The LISTING PRINCIPLE subsumes the means for enumerating in sequence a set of related or comparable items. Like the other six linear principles, this one applies to all scales and levels. On a local scale, the list may be inside one sentence:  

 (190) The Mechanical Engineering department offers the following degrees: Bachelor of Sciences in Mechanical Engineering, Master of Engineering, Master of Science, and Doctor of Philosophy.  

List items may fill one paragraph each, as in samples (511, 515, 516, and 541) in VI.2.18, 20f, and 33. On a still more global scale, items may each fill a series of paragraphs, as in the discursive listing of the seven principles of linear operation in this book section.

   2.70 Formally, listing is a type of RECURSION. One category in the linear sequence is selected and repeated. Processing adapts by re-activating its operations for each recursion until the list terminates — a specialized use of look-back and look-ahead (cf. IV2.35ff). This tactic cuts down on load so that the normal preference for twos and threes can be suspended without strain, e.g., for the four direct objects in (190). Presumably, processing specialists identify the shared criteria of the list and integrate as many items as required (IV.2.13). Look-ahead dominates if the criteria are evident at once, and look-back dominates if they must be extracted after consulting all or most of the list. List processing would function best if the list items facilitate these operations. Lists should be controlled for unity, explicitness, and ordering.

    2.71 In a unified list, the items are distinctly comparable among themselves. On the conceptual level, the items should have clear content relations, as in (190) (‘degrees’). This unity should be recognizable after two or three items at the latest, so that processing can adapt accordingly (IV.2.70). Conceptual relations decide if a list is disjunctive, e.g. (191), or additive, e.g. (191a) (cf. IV.2.72):  

(191) young, middle, and old adults (Meyer, Rice, Knight, & Jensen, 1979: 1)

(191a) young healthy, and energetic adults  

Mutually exclusive concepts sort the list into as many groups as items (191), while compatible concepts yield one group with multiple attributes (191a). A markedly disunified list should be disturbing or surprising, e.g., in order to confuse an audience:  

(192) “The time has come,” the Walrus said, “to talk of many things: Of shoes — and ships — and scaling wax — of cabbages — and kings” (Carroll, 1960: 235)

Conversely, list items too close in meaning appear redundant: {190}  

(193) Quarterbacks need quickness, speed, fast reactions, and rapid motions  

except for special effect. e.g., to highlight repeated actions:  

(194) His mind was always on the script, whether he was sitting in a bar, driving his car, sitting in a bar, relaxing at home, sitting in a bar, eating at a pizzeria, or maybe even sitting in a bar. (Burns, 1976: 175)  

On the phrasing level, parallelism supports unity. The items on the list may be all cores, e.g. direct objects (193) or verbs in the predicate (194, 196a); or all adjuncts, e.g. adjectives (191, 191a). In (190), the list names ‘degrees in mechanical engineering’ formatted as noun phrases (noun + preposition ‘of’ + noun) in apposition with the direct object of the main clause. Nonparallel phrasing of lists in unskilled writing can be a needless processing obstacle. In (195), from a paper on juvenile delinquency, the student obscured the boundaries of list items, whereas her revision (195a) is much clearer:  

(195) The child must do one of several things — pay tuition or a fine, community service, or even a suspended disposition.

(195a) The child must do one of several things — pay tuition or a fine, perform community service, or even accept a suspended disposition.  

Recurrences and pronouns can be saved by making a compact, parallel list. Contrast (196) with its revision in (196a):  

(196) The architectural engineer studies the drawings and plans. He inspects the building to insure that it will withstand everyday use. The engineer looks for ways to minimize the use of energy and space. He makes suggestions to the architect.

(196a) The architectural engineer studies the drawings and plans; inspects the building to ensure that it will withstand everyday use; looks for ways to minimize the use of energy and space; and makes suggestions to the architect.  

     2.72 In an explicit list, the category to which items belong is announced. A colon or typical listing signals can be inserted (e.g., ‘the following’, ‘includes’, ‘first’/’second’/’third’, etc.). It is strategic to place the list late in a sentence, where audiences expect focussed materials (IV.2.63). Therefore, the announcement typically precedes the list, as in (190) and (195), so that look-ahead can begin at once (IV2.70). Less often, the announcement comes after:  

(197) Bookkeeping, games, and communication are three common uses of micro-computers. Either way, the final list item is usually introduced by a junctive: ‘and’ if the list is additive, and ‘or’ if the list is disjunctive (IV2.52). Odd placement of junctives confuses the list (cf. IV3.38).  

    2.73 An ordered list is one whose linear succession has a discoverable motive, e.g., moving forward in time:  

(198) The contractor builds the foundations, sets up the walls, and puts on the roof.  

or progressing to steadily greater degrees of heaviness:  

(199) Possible options include diplomatic protest, economic sanction, military pressure, or declared war.  

These ‘options’ go from less to more extreme—the usual sequence for building gradually to a climax (cf. VI.2.34), e.g., in an international confrontation. Here, listing interacts with heaviness as a sequential strategy. The reverse seems oddly anti-climactic, as in this student passage about the hostages in Iran:  

(200) They experienced fake executions, beatings, imprisonment, isolation, and uncertain futures.  

    2.74 The listing of modifiers is a special problem (cf. V.2.14). The list is often linearly fixed, e.g. in (Halliday & Hasan, 1976: 40): 

(201) two high grey stone walls along the roadside  

Yet the ordering criteria are hardly defined and certainly not learned in school grammars (cf. Ziff, 1960; Vendler, 1968; Martin, 1969; Danks  & Glucksberg, 1971). Quirk et al. (1972: 924ff) suggest:  

Evaluative or subjective adjectives frequently precede those relating to size, while these in turn frequently proceed those of girth; […] modifiers relating to properties which are (relatively) inherent in the head of the noun phrase, visually observable, objectively recognizable or assessable will tend to be placed nearer the head and be preceded by modifiers concerned with what is relatively a matter of opinion, imposed on the head by the observer, not visually observed, and only subjectively assessable; [but] there is plenty of room for difference of opinion.  

Similarly, Judith Martin (1969) argued that the more “instrinsic” modifiers—closely tied to the nature of the object/event expressed via the noun head—are placed nearest the head and selected earliest (with the aid of spreading activation?), though uttered last; but experimental evidence is inconclusive. Young children may have trouble with modifier sequences (cf. V.2.12f), e.g. this sixth-grader who wrote  

   (202) The fast, enemy, slow airplane was flown by the intelligent, killer, tired robot.  

defying both cohesion and coherence.

    2.75 The seven principles of linear processing have now been illustrated on various scales and levels. The seven regularly interact: cores vs. adjuncts being separated by pauses, or determined by heaviness; look-back balanced against look-ahead; lists controlled by heaviness or look-ahead, and so forth. Under heavy load, the processing system might be forced to sacrifice some principles. Disambiguation is commonly neglected when texts are produced in haste, e.g., newspapers rushing to press (IV2.68). Extremely disordered speech may reduce core-adjunct patterns toward lists, such as the “telegraph style” of sonic aphasics (Luria, 1979: 285).¹[ Luria’s (1979: 285) illustration was: ‘Oh / the jackdaw / the doves / good food / the white color / deception / a cry’. On an interesting use of telegram style for a special goal, see V.2.32.] Look-ahead seems to degrade before look-back (IV.2.47, 50f). Heaviness shifts from later to earlier sentence positions if the producer is strongly involved (IV2.64). In any case, the modality of writing is better equipped to control, reconsider, and revise linear patterns than speech is. Most possible problems are ruled out by punctuation: a system for marking linear relations and thereby guiding processing. In section IV.3, I shall try to show how punctuation fulfills that function.

Click here to go to Chapter IV, Part 2