Version classiqueVersion mobile

Text and Genre in Reconstruction

Willard McCarty

2. Cybertextuality by the Numbers

Ian Lancashire

Texte intégral

1When we think in words, the thoughts come in grammatical form with subject, verb, object and modifying clauses falling into place without our having the slightest perception of how the sentence structure is produced (Lashley 1958).

2I suppose that all of us have a primitive prompter or commentator within, who from our earliest years has been advising us, telling us what the real world is. There is such a commentator in me. I have to prepare the ground for him. From this source come words, phrases, syllables; sometimes only sounds, which I try to interpret, sometimes whole paragraphs, fully punctuated. When E. M. Forster said, ’How do I know what I think until I see what I say?’ he was perhaps referring to his own prompter (Bellow 2006: 95).

1. Introduction

  • 1 The works of Sigmund Freud make the acceptance of this simple truth much easier. As Karl Lashley s (...)
  • 2 See Lancashire 2004b. Cybertextuality extends Espen Aarseth’s term ‘cybertext’ (1997), which is fi (...)

3Readers who want to find the author in the text should not be too disheartened when they see only themselves. The New Criticism and Reader Response theory gave up trying, not unreasonably, because authoring takes place largely in the author’s unconscious and resists being observed.1 Cybertextuality,2 the subject of this essay, theorizes the authorial process by bringing together observations from four areas: Norbert Wiener’s cybernetics, self-testimony by creative writers, cognitive psychology, and computer-assisted text-analysis. I first apply Wiener’s communications theory to describe how authors craft sentences by subjecting their own chunk-sized phrases to mental cycles of uttering (messaging) and error-monitoring (selffeedback). Cybertextuality pays close attention to evidence for two hypothetical cognitive quantities, the cognitive chunk or phrasal word-packet, and the cognitive capacity or load. I then review what cognitive psychology knows of these numbers, and of our mental self-monitor.

4I refer to chunk size as the alpha value. What constrains it is the capacity of the phonological loop of Alan Baddeley’s model of working memory (Baddley 2000). This constraint comes with being human, although an author can appear to get around it by shuttling chunks rapidly in and out of working memory from the clusters or schemas that form, distinctively, in long-term memory networks and serve domains of knowledge in which the author has expertise. Computer text-analysis (stylistic counts and the repeating patterns in textual concordances) detects what even the author cannot perceive about himself: it reveals these chunks in the phrasal vocabularies of authors (Lancashire 2004a). Chunk size also limits the error-finding capacity of our mental self-monitor: mistakes that cros chunk boundaries normally escape notice. Authorial self-repairs in a holograph manuscript, such as Shakespeare’s hand in The Book of Sir Thomas More, (1910) show this mental feedback at work. The second constraint is the omega value, the cognitive capacity or load for a thought. Virginia Woolf’s holograph manuscript versions for The Waves (1976) reveal this size-constraint for her mind.

  • 3 I call them alpha and omega because, in operating within their constraints, our minds approach as (...)

5The alpha and omega values, these cybertextual numbers3, inform the authoring process and help shape an author’s idiolect. They do not change markedly from person to person. These two verbal capacities – of chunks, and of maximum text-size that chunks may combine to express a unified thought – also drive authors, both to invent language technologies that supplement their memory capacities and to empower conscious self-editing. These technologies act through the laws of copyright to replace the concept of a creative author by that of an owner-author. The post-literate cyborgic author is a self-employed expert at expressing (not thinking) verbally (not conceptually) original texts that are always fixed in some medium. Fixation, when utterance takes an external form, is the critical step. It marks where the author-creator gives way to the author-owner, who may even be defined as someone who has never even read the authored text he owns.

2. Cybertextuality

6Cybertextuality theorizes our authoring of speech and written text, in the context of what we know about language cognition and about the technologies by which we supplement that. Every language tool that we have made, from oral-formulaic metre to the digital workstation, speaks to our species’ frustrations with native cognition in language production, that is, unassisted speaking, writing, and reading. Dissatisfied with how genes have designed us, we have built communications technologies that give us a flawlessly searchable and readable external long-term memory and an aid to working memory by which we can author error-free, carefully sculpted utterances, both brief and epic. The author uses a literary machine (this is Ted Nelson’s term (1987), modified in Katherine Hayles’ Writing Machines (2002)) as a cybernetic extension of her mind.

  • 4 Mysak specified ten loops, two of them feedforward (ff) and five of them feedback (fb): thought pr (...)

7The cognitive mechanics of authoring need cybernetic concepts: both Espen Aarseth’s book Cybertext (1997), which describes texts (such as interactive fictions) that ask readers to do physical work, and posthumanist Katherine Hayles (1999), stimulated theory to re-think Norbert Wiener’s communications theory. The human-computer interface supplements and imitates cognitive language production by the inner Muse and her receiver, the inner Editor. We have even developed text-analysis software that facilitates the reading of texts. Such tools extend the human mind in uttering. By availing ourselves of them, we become cyborgic and partake of the character of a cybernetic organism. Authoring of texts is a recursive process in which hand-shakes cycle between a sender who utters something and a receiver who perceives the sent message and feeds back information about it to the sender. Unlike Wiener’s cybernetics, cybertextuality asserts that the sender and the receiver are initially the same: we read ourselves before anyone else does. This is an old idea found in the speech-pathology research of Edward D. Mysak (1966). He argued that the ’speech system may be viewed as a closed, multiple-loop system containing feedforward and feedback internal and external loops’ (1966: 17), that is, loops that take place wholly within the mind, as distinguished from Wienerian loops that involve the reactions of other listeners.4 Cognitively, everyone utters speech, syllable by syllable, only after extensive self-monitoring for errors, and then self-adjusting for corrections.

8Wiener coined the word cybernetics in the 1950s to name a theory of communication that has since morphed into information science and has influenced how everyone thinks (Wiener 1950/1967). The word comes from the Greek word for steersman, kubernetes. Cybernetics asserts the commonalties between how people communicate with people, how machines such as anti-aircraft guns and radar technology signal one another, and of course how people negotiate with machines like text-analysis tools. Wiener had the brilliant idea that communication is a hand-shaking exercise, a series of message-response transactions between sender and receiver. Each sent message prompts a response that enables the sender to correct the first message or to move on to something new. Wiener’s five cybernetic modules are sender (author), receiver (reader), message (text), channel (medium), degrading noise (language and culture change, and more), and feedback (criticism). His ’theory of messages,’ as he referred to it (Wiener 1950: 106; Masani 1990: 251-52), is that the sender steers or controls composition according to a receiver’s incremental feedback because only after that response can the sender be certain that noise has not rendered the message unreadable. A message is thus an incomplete act of communication without its partnering feedback.

9Biology has for a long time been influenced by cybernetics, and cognitive scientists frequently describe mental and neural processes as message-feedback in nature. Wiener’s universal phenomenon, in cybertextuality theory, also embraces many aspects of literary research. For example, rhetoric and genre can be conceived as strategies of redundancy employed by authors who want to reduce reader misunderstanding that arises from cultural, gender, and social differences as well as from transmission errors often identified in textual criticism and scholarly editing. (Redundancy sacrifices cognitive message size to make messages immune to disruption by noise.) Reader-response theory internalizes the message-feedback partnership in the reader rather than the author. The scepticism that moves postmodern theorists to speak of the death of the author arises from their legitimate perception of how unconscious the author is of cognitive creative process, of how mysterious it is. We retreat into Michel Foucault’s author-function (Foucault 1989/1969), that is, the owner-author in which law and capitalism trade.

10From a cybertextual perspective, however, the author is partly alive in the work. Messaging and feedback cycles leave idiolectal and idiosyncratic traces that survive the author’s replacement by the editor, the owner-author, the employer who asserts property rights on an authoring employee, the buyer, death itself, the deceased’s estate manager, and especially the reader. Texts incorporate fossils of their own biological realization. Cybertextuality asks us to read texts in a new way, to discover within them the stigmata of authoring − the marks that distinguish its subjection to cognitive limitations. These are partly observed in uttering-feedback cycles as they create evanescent or frozen texts.

  • 5 A researcher reading a blurred photocopy of an article, or someone listening to a friend’s goodbye (...)

11Of course, Norbert Wiener’s five cybernetic modules do not map onto authoring as neatly as one might like.5 The author is apparently unknowable, for the time being at least a see-through cellophane entity that we cannot perceive directly because we have no memory system to store (and thus consciously verbalize) what it is. The reader is the only part of the author whom we can see at work, subvocally as error-catching and inner speech, externally as what our writing and speaking tools utter back. Unlike Wiener’s messages, both uttering and feedback are sent as phrasal fragments. Sometimes the sender fashions an utterance for two or more receivers and crafts the message so that they find different things in it. Messages are readable first, not externally, but in the sender’s own working memory, a little living auditory and visual book that dynamically overwrites itself in seconds. Then they appear in iconic or echoic sensory memory for a brief time, unless machine-assisted. The channel for these cybertextual transactions is the mind’s neural infrastructure, the brain, but the brain is not the simple left-to-right pipe by which Wiener’s cybernetics represents the arena across which a message travels from sender to receiver. For one thing, as lesions prove, the channel can break down and generate the cybernetic noise that damages transmission. Uttering uses not a single channel but a crosshatch of dynamic and parallel processes operating and interacting simultaneously. Only when an utterance reaches our sincerest flattery of the brain, the computer, does it take a simpler path. The sources of noise that afflict cognitive messages in transmission – dementia among them − also resist easy explanation, although imaging devices, and biopsies and autopsies, have for a century gradually unveiled the havoc that cerebral lesions can wreak on cognitive functions.

12Cybertextuality asks cognition patently awkward or impossible questions about an imprecise phenomenon, literary making. Scientific consensus abhors the quick answers we would like, and even science changes its mind sometimes. Cognitive-psychology experiments focus tightly on testing hypotheses advanced by previous experiments. Research runs counter to the grain of basic science when it treats governing theories, like Alan Baddeley’s working memory, Willem Levelt’s lexical access in speech production, and Walter Kintsch’s construction-integration theory of text comprehension, as other than explanations that are consistent with cognitive affects and other experimental results. Not infrequently, an explanation for a well-attested empirical effect, accepted for some time, suddenly is controverted by a related experiment. Yet even cybertextuality theory, while speculative, anchors itself in an experimental science, text analysis. Here, the subjects of text-analysis experiments are Shakespeare’s hand in the manuscript play of Sir Thomas More, and Virginia Woolf’s method in creating The Waves.

3. Cognitive Authoring and the Chunk

  • 6 See Geschwind (1979), a classical account of how the brain handles language; and Damasio (1994) an (...)

13Most of the human brain, topographically, is active at some point during an utterance.6 The purpose, the plan, and the gist of a message originate in the pre-frontal cortex, like every other thing we intend to do. If we decide to read a sentence aloud, for example, processing begins once the visual cortex at the very back of the brain receives data from the eyes. The frontal cortex then sets in train an acoustic re-encoding of this visual data so that they assume the form they would have had if heard and passed on by the temporal auditory cortex. Next the brain must perceive semantically what is said. Wernicke’s area in the left hemisphere in the brain of a right-handed person is associated with semantic comprehension. This assigns words (lemmas) to each sound from long-term memory. If Wernicke’s area is damaged, the brain can produce word-salad sentences: they have a proper syntactic form but do not make sense. Whatever emotional (rather than logical and linguistic) content a sentence has is recognized by the right hemisphere of this right-handed person. Then the brain encodes the recognized word-train syntactically and phonologically: Broca’s area, just forward of Wernicke’s area, does this. A brain in which Broca’s area is damaged utters semantically understandable sentence fragments lacking grammatical form. Activities at both sites tend to involve areas immediately under the cortex. Finally, the fully-encoded sentence moves to the motor cortex for pronouncing. Once the brain gives phonological form to an utterance, it can enter the catchment area of working memory as inner speech. After semantic and affective processing, we hear the sentence unfold subvocally and overtly aloud. It never occurs to many people that language production is unconscious because we experience its assured power immediately in uttering speech.

14Functionally, this process takes one step after another, but it very probably happens all at the same time, each sub-process affecting every other. Different aspects of uttering take different paths through the brain. If we type an invented sentence, for example, the brain must re-encode the auditory data native to its language function into visual form and must issue instructions to the motor cortex on how to operate a keyboard. Localization of brain functions thus reveals the modularity of mental language processing. Certain activities take place in specific places in the brain, even though many activities are active simultaneously. Localization has proved especially meaningful in analyzing the writing of individuals who have been subject to medical testing.

15Peter Garrard’s research on novels by Iris Murdoch as she succumbed to Alzheimer’s disease – an essay found elsewhere in this volume – is a uniquely important case study in that respect. A ’profound bilateral hippocampal shrinkage’ that turned up at an autopsy after her death in 1999 affected the linking of her working memory to her long-term memory. Jackson’s Dilemma, her final novel, exhibits lexical and semantic impoverishment, with a drop of twenty percent in vocabulary. This result was not unexpected: Alzheimer’s Disease sufferers have a ’relative impairment in semantic processing’ (Poore and others 2006). The medical tests and autopsy had turned up a medical condition – in effect, a growing deterioration in one location of the brain, something like a lesion – that coincided with a reduction in Murdoch’s lexical store and in the information density of her prose. The link between Murdoch’s body and mind in Garrard’s case study was her working memory. The hippocampus, its gateway, must also be involved with the most important part of our mental life, where we directly and self-consciously hear and manipulate our language subvocally. When we hear ourselves think in the mind, we use working memory to do so. Sufferers of dementia gradually feel that space shrink. If there is any part of the complex cybertextual channel that we can access directly, that part is our working memory.

16Alan Baddeley’s sturdy model of working memory (formerly called short-term memory) has lasted nearly thirty years (Baddeley 1986). It has four functions: the phonological loop (where we hear subvocal mental language), the short-term visual sketchpad (where we summon up or store images, including written text), the episodic buffer (where we retrieve multimodal episodic memories), and an executive that manages these three slave systems. We use the executive to refresh what appears in either store, or to overwrite what is there with something else. Because our brain can only manage mental language if it is auditorily encoded, the visual sketchpad can keep a printed or a written sentence or page before our consciousness as an image, but before we can manipulate text cognitively as language we have to re-encode it phonetically (read it); thus we transfer it into the phonological loop. That loop, a kind of cognitive mobius strip, imposes big constraints on our linguistic self-awareness. The most important limit is on how much spoken language we can place in and retrieve from working memory. What can we say about its size in a healthy person? I refer to this size the alpha value because it characterizes where every utterance starts.

17One of the best known quantities of working memory is George Miller’s ’magical number’ (1956): this is the maximum number of elements we can hold in active memory – seven, plus or minus two elements. Telephone numbers were allotted seven numbers for this reason half a century ago. Now that we have an area code as well, another three digits, we can only consciously keep a phone number in memory until we dial it by making several chunks represent two or more numbers. In 2000, Miller’s ’magical number’ was revised (on the basis of decades of experiments) by Nelson Cowan downwards to four chunks, plus or minus two chunks. Cowan estimates that the still uncertain size of each chunk is perhaps three or four items. Another variable is the Brown (1958) / Peterson (1959) duration rate for short-term memory items, which decreases markedly until it almost reaches nil at twenty seconds. An equation for a person’s working-memory retrievability capacity, therefore, would have at least three variables: chunks, items (or subchunks), and their current duration, measured in seconds, within working memory. These quantities have been proposed by cognitive scientists after dozens of experiments with human subjects over half a century. Arranged in my simple model formula, they model one’s retrievability capacity, the number of potentially retrievable items from Alan Baddeley’s phonological loop, at any given point in time, as the alpha value.

  • 7 I hope that someone else, more qualified than myself, has devised or will devise a better formula.

18The formula is α = (β * γ) * δ, where beta (β) is Cowan’s chunk capacity, a number from two to six; gamma (γ) is Cowan’s items per chunk, a number from three to four; and delta (δ) is the percentage of items still retrievable, measured by the seconds so far held in working memory since the last refresh or rehearsal, according to Brown and Peterson.7 They estimate that memory loss increases, the longer the duration: nothing lost after no seconds, fifty percent remains after three seconds, forty percent after six seconds, twenty percent after nine seconds, twelve percent after twelve seconds, and so on. The range of alpha values goes from 0.6 items, for someone with the lowest chunk and item capacity near the end of maximum duration in working memory, to 2.8 items, for someone with a median chunk and item capacity in mid-duration, and to 24.0 items, for someone with the highest chunk and item capacity at the very point when everything is planted or refreshed in working memory. Fernard Gobet and Gary Clarkson (2004) demonstrate that, for experts (those with the highest capacity), the number of chunks is fewer than three, but they are templates from longterm memory and the items they hold are go up to fifteen. The Gobet and Clarkson alpha value, then, for an expert with fresh working-memory content would be (2 * 15) * 1 = 30 items. The inner voice would have, if Cowan’s experiments are definitive, half a word at the least, three words on average, and (at the most) two alexandrines or 12-syllable lines (or three pentameter lines, following Gobet and Clarkson).

19The alpha value does not give information calculated in bits. That amount is less than the items represented by the alpha value because all human languages are redundant. Further, the alpha value is not an intelligence quotient, or a measure of long-term memory size, or a limit on what we can communicate to others. Almost all speech and thought-making take place in the unconscious mind and feed off a long-term memory store that we can probe but never read. Damage to phonological working-memory capacity does not impede sentence comprehension (Martin 1987). Alan Baddeley explains that, on the basis of experimental work with children and secondlanguage learners, the phonological loop is the system that evolution has developed for the crucial task of language acquisition. Adults who have a disruption to this system do not have too many problems, provided they are not required to learn new languages (2004: 54).

20We are aware of an intelligence working in the cognitive background as a gist of something we want to say. Although we can consciously schematize the process of thinking, we cannot recall how this gist emerges, that is, how we make a thought. Working memory thus limits the mastery of language, not thought. It is the mental function (other than immediate auditory or echoic memory) where we are conscious of what our mind generates or receives before vocalization and where we can do something about that awareness, such as commit it to long-term memory or to paper. Our unconscious continues to feed language effortlessly, spontaneously, into subvocal working memory in chunks that fit within the phonological loop.

21The retrievability limit represented by the alpha value is unaltered by what enters our sensory stores, whether echoic or iconic. Both these immediate forms of sensory memory, echoic or auditory, iconic or visual, decay rapidly after 250 milliseconds or a quarter of a second and are gone utterly after half a second (Robinson-Riegler 2004: 104-5). The longer we continue to look at something, oddly, the more the iconic image decays (Robinson2. Riegler 2004: 106-07). We can deliberately store what we see in sensory experience in the visual sketchpad, but only after conversion to phonetic encoding can we manage it as language. As a result, we again appear to bump into the constraints of the phonological loop. What we hear from others as vocal speech overwrites some or all its contents. That is why it can be hard to finish a thought in words when someone suddenly breaks one’s attention by talking to one. Our phonetic working memory has no shut-off valve, no filter, to save us from an incoming utterance in a natural language we know.

4. Cognitive Reading

22It’s unclear whether we actually edit what we say before we say it, but the fact that we edit what we have already said is not in doubt (Robinson- Reigler 2004: 421).

23From time to time we catch ourselves in making a speech error. We stop what we are saying in mid-flow with an editing comment such as ’uh, that is’ or ’um, I mean,’ and we substitute a corrected word or phrase. These spontaneous self-repairs illustrate cybertextual cycling. (Comparable are the kinds of word-processing corrections that we make as we type a sentence.) Hesitations, pauses, or rests are another sign of message-feedback cycling. We may not know why we are hesitating after one phrase, and before the next, except that we are not ready to move on. In oral delivery, we nervously use fillers or paralanguage to keep the listeners’ attention and signal that something else is on its way. If we time a lecturer’s use of these fillers, one every few seconds, we can see chunking and cybertextual cycling at work.

  • 8 Baddeley suggests that the auditory imagery system, not working memory, may be responsible for inn (...)

24Self-monitoring also occurs at the pre-vocal cognitive level when a mistake is interrupted in the middle of uttering it. Willem Levelt (1989: 467-8) proposes that, at different levels of utterance-processing (lexical, phonological, etc.), ’a watchful little homunculus’ monitors, editor-like, ’the construction of the preverbal message, the appropriateness of lexical access, the well-formedness of syntax, or the flawlessness of phonological-form access.’ His self-monitor intercepts or receives both inner (subvocal) and covert speech, checks it for errors, and reports back to what he calls the ’conceptualiser’ so that it can reformulate the utterance (cf. Hartsuiker and others 2005: 4). As an example of an error caught subvocally, before articulation, Levelt gives the example, ’we can go straight to the ye-[...] to the orange dot,’ and explains that ’To interrupt right after the first syllable [ye-], the error must already have been detected a bit earlier, probably before the onset of articulation’ (1999: 33; and for additional evidence, see Hartsuiker and Kolk 2001). Another recent experimental finding, from magnetoencephalography, is that self-uttered speech takes about 100 milliseconds to activate the speaker’s own auditory cortex (Curio and others 2000: 190). This suggests to me that self-monitoring of such errors occurs in prevocalic inner speech,8 somewhat in advance of actual vocalization. Levelt believes that self-monitoring can only be effected by a cognitive process parallel to the one that authors an utterance. This process would read incremental stages of composed utterance and return feedback to the authoring process. Both processes would be simultaneously active and, at each exchange, would effect a cybertextual cycle.

25There is already reason to believe that we cognitively model or construct what we hear and see, accepting our own model over the initial sensory data. Philip Lieberman (2000: 57, citing McGurk and MacDonald 1976), explains the well-known McGurk effect. When a person sees a video of the face of someone saying ga at the same time as the video sound system outputs ba, that person hears the intermediate sound da. If the person covers his eyes, however, he hears ba. Test subjects, faced with an inconsistency in the data (what might be termed noise), unselfconsciously hear only what they analyze as being said. The McGurk effect reveals that we use visual evidence to determine what a sound is, but it also demonstrates that we integrate, in a mandatory and spontaneous way, audiovisual speech syllables that disagree with one another and create a sound never made. Posner and Raichle report another instance of how a deep cognitive process emends sensory input: ’if one removes a phoneme from an auditory word and replaces it with white noise, what is often heard is the correct word with a burst of noise superimposed’ (1994: 112). Semir Zeki agrees that ’One of the functions of the brain... is to instill meaning into this world, into the signals that it receives,’ but he cautions that often the mind must ’allow of several interpretations, all of equal validity.’ (2006: 262). We experience these alternate meanings sequentially and do not feel obliged to select among them.

26The brain betrays its cybertextual cycles in error recognition and recovery. Self-monitoring may extend to any cognitive language representation, whether in uttering or reading a word or syllable, where error or ambiguity is perceived. For example, the remarkable N400 brain waves discovered in electroencephalographs (EEGs) of individuals reading semantically problematic or unexpected words show that the long-term memory store reads the message as it unfolds and comments on it using data that appear to be non-linguistic in nature. An N400, negative-voltage wave in the brain peaks 400 milliseconds after it encounters a semantically incongruous word (Robinson-Reigler 2004: 391; Federmeier and Kutas 1999). The N400 wave registers a channel-wide response to an unexpected mental utterance such as ’Experience is the mother of despair.’ After formulating a message partially, the brain responds to it, occasionally with a N400 wave, other times with vocal self-repair. Because we cannot script our words haltingly in working memory before we say them, because we seldom know exactly what we are going to say until we actually say it subvocally or aloud (as E. M. Forster says), and because we hear ourselves speak and see ourselves write at almost the same time as a listener or a viewer does, we need an unselfconscious error-checking facility in place to catch mistakes.

  • 9 We fixate more briefly on function words than on content words (Gleason and Ratner 1998: fig. 5.4)

27Most speech errors caught by our cognitive self-monitoring involve a switching of word-onset consonants in small phrases (e.g. phrall smases). These are well within the alpha value for management in working memory. Larger errors in syntactical structure, such as subject-verb agreement in number, elude detection until we engage in close editing and proofreading or run an analysis by automatic style-checking software. The perceptual span in reading a text has a comparable capacity to the phonological loop. Our eyes traverse in successive fixations, saccades (left-to-right jumps), and regressions (reverse saccades). A saccade takes twenty milliseconds and traverses six to eight letters, and a fixation lasts 200-300 milliseconds,9 unless it settles into a gaze, and encompasses about three characters to the right and fifteen to the left (or four to five words in length). College-level students move and fix their eyes 90 times for every 100 words, 25 percent of which saccades are regressive (Crowder 1992: table 2.1). The phrase, ’watchful little homunculus,’ describes both the cognitive self-monitor and the reading eye. Both perceive mental speech segmented in comparable spans whose capacity may approximate also the alpha value.

  • 10 I discuss the omega value for this below.

28Language self-consciousness appears to be a stream but, when examined closely, consists of staccato-like pulses in which a succession of chunks, proposed by a cognitive conceptualizer (which we experience as the gist of what we intend to say), are monitored for correctness by a parallel process before being articulated. The Muse who brings texts piecemeal into being from darkness, and the Editor who announces corrections to those texts and knits them together from much the same obscurity, feedforward and feedback our utterances in cybertextual cycles. Messages are normally fragments. The phrasal size of caught speech errors, and the perceptual span in reading, both fall near the alpha value. Is this value, then, a broader cognitive storage limit? Although the brain has ’1011 neurons connected by 1015 synapses’ (Chklovskii and others 2004: 782), and the consensus is that our long-term memory capacity is limitless, working-memory capacity is conceded to filter and constrain information, especially language data, entering the long-term store. Is it possible, then, that there may be a maximum conceptual span, an upper limit on the extent of a cognitively manageable associative memory cluster?10 Would it be so odd if all memory systems worked by the same numbers?

29Gilgamesh, the Bible, War and Peace, and La Divina Commedia are not just phrasal sequences: these epic achievements have, camel-like, come through the eye of such needles as. It is unclear whether a faulty phonological loop would have prevented Dante, Tolstoy, and Shakespeare from writing. However, because language is always changing, language learning never stops. If evolution designed the phonological loop for language learning, then it is also, to a degree, the Procrustean bed of language. Has anyone in recent memory created a substantial utterance, novel-sized, simply with the unaided human memory system? Most of us cannot manage a list of groceries, even though we may well be able to use language well enough not to starve.

30How have we overcome the limits of these values? We have externalized the Muse and the Editor. Writing, printing, and digital tools have exported much of our Muse, that is, our long-term memory and lexical production system (as Levelt calls it), into dictionaries, encyclopedias, and the world’s vast libraries. Word processors operating on computer workstations with visual displays have imported some of the knowledge by which Willem Levelt’s ’watchful little homunculus’ corrects speech errors. Our workstation adds two additional dynamic systems to those which that our cognition calls its own. While the eyes are fixed on the page and on the screen, iconic memory maintains a much fuller copy of what we have uttered than working memory could ever have stored.

5. Cognitive Capacity

31What impact does a thought-capacity limit – the omega value – have on our uttered texts? Except for conversation, where the amount of text between turns is quite small, we do much better than generate phrases and paragraphs. The process by which we cobble our chunks into speeches, essays, and books uses two technologies, the more recent being external memory and processing systems from writing (which first unfettered length restrictions on sentences and paragraphs) to word processing, and a second, more ancient one in the loci et imagines method of artificial memory described in Frances Yates’ The Art of Memory (1966). It asks the extemporaneous speaker to prepare by mapping the order of his sentences as stops on a walk through a public place like a forum or a mall, and the content of each of those sentences in an object placed, mentally, in each stop. As the author speaks, he guides himself by mentally taking his own guided tour through the place, pausing at each stop to pick up its object. (For example, if I memorized the structure of this essay by mapping it to my local retail park, I might think of artificial memory as the local Indigo bookstore and place a toy locomotive, symbolizing the loci et imagines method, at its front door.)

32This method can be observed today in the behaviour of experts and mnemonists. Chess masters can rapidly move schemas (the templates in Gobet and Clarkson 2004), which are specially encoded and stored in long-term memory, in and out of working memory. They can expand the sub-parts of these schemas without losing track of where they are. This mental technique shuttles between the two memory systems, leading to a ’10-fold increase in performance on tests of STM [Short Term Memory],’ but only in their specific areas of expertise (Ericsson and Kintsch 1995: 211-12). Normally it takes between five and ten seconds to store a memory long-term and a second to retrieve it, but experts manage to make retrieval from long-term memory only 300 milliseconds longer than from working memory (1995: 215). This technique draws not on deductive logic but on stored memories of sequences of moves and positions on a chessboard. Experienced writers and readers have similar stocks of sentence and paragraph structures in long-term memory that can be applied at need. Writers like John Milton, Henry James, James Joyce, and Cormac McCarthy can produce very long sentences, extending well beyond page boundaries. Periodic styles like Milton’s in Paradise Lost, composed mentally in daily segments while he was blind, and dictated to his daughters in the morning, must have grown from decades of expertise as a master of Latin, Greek, and Hebrew, and as Oliver Cromwell’s Latin secretary. Milton and these other writers are as great experts in English composition as chess grand masters are in positional play. The making and exploitation of these schemas constitute individual traits, but they do not enlarge personal cognitive capacity. The expert learns to associate a more-or-less abstract analytic map in long-term memory with many small but fully-realized codes for that map in working memory.

  • 11 XML (Extensible Markup Language) is a set of rules for encoding documents electronically.

33That we cannot use an expert’s mental technique generally suggests that some general cognitive capacity limit exists on our ability to comprehend thoughts as well as utterances. Edward Sapir and Benjamin Lee Whorf believe that we can only think in natural language − to them, there is no distinction between thought and language − while others believe that we think in non-linguistic concepts, often called mentalese, and then translate its output into language (Pinker 1994: 67-82). Language being a late evolutionary development, should we deny thought to so many other languageless species (including our own, not so long ago)? Insofar as language represents only a part of the content of our long-term memory, also, does it make sense that we do not think with the images, events, and sounds stored in it when so many testify otherwise? A non-linguistic mentalese offers a plausible foundation for the making of all languages from Swahili to symbolic logic and music, much as XML,11 for example, enables us to devise a multitude of encoding languages. The inchoate gist we feel before saying something has not been shown to be natural language.

  • 12 The Brown Corpus is a carefully compiled selection of current American English, totalling about a (...)

34If languageless thought observes a cognitive capacity limit, does it exceed the alpha limit on speech in working memory? Well, of course who cannot comprehend instantly a sentence that takes longer than two seconds to utter? That granted, a limit manifests itself in many ways. When we artificially interfere with someone’s working memory, as by giving a writer a second, simultaneous task, we know that his sentence length falls. We know that college students who had to hold in memory six digits concurrently even as they devised two-noun sentences wrote significantly shorter sentences (Kellogg 2004). The practice of professional writers is also telling. Average sentence length in 58 articles in The Independent newspaper is only 24.58 words (Hearle 2007), that is, about two alexandrines in length. Oxford University Press recommends that sentences average 15-20 words. The most frequent length for sentences in the million-word Brown Corpus12 is twelve words (Sigurd, and others 2004). Sentence length also turns out to be a factor in syntactic difficulty, as vocabulary length (in word-syllables, for example) is in semantic difficulty: together those lengths define readability measures like the Flesch-Kincaid (Kincaid and others 1975). If our minds did not have a cognitive capacity limit during reading, we would not have to develop readability measures that often exceed the phonological loop. These measures imply that a capacity limit exist for understanding outside working memory as well as for phonemes inside it. How big is this limit, and how do we even find the words and the measures to tell?

35The omega value here names how large a language-mediated thought we can mentally grasp. This conceptual span is related to the gist with which authors begin in formulating an utterance. If we reverse the cognitive sentence-production process, translating sentences back into concepts, as readers do in text comprehension, we reduce an utterance with many parts (syntactic structures that build on phrases and words, supplemented by information on how to pronounce) to an originating, graspable thought. So far, we have not been able to measure how large this conceptually graspable entity may be, but several approaches look promising. Two appear in cognitive psychology: our ability to measure the psychological status of propositions; and the discovery that students who can solve a problem by using a worked-example fail – by exceeding their cognitive load – when asked to employ an ends-means technique. Others belong to the humanities. For example, self-imposed text-length restrictions plausibly register sensitivity to a general human cognitive capacity, so do the size limits of poetic forms. So far, little attention has been paid to text genres as evidence of cognitive capacity.

36Propositions in cognitive psychology are subject-predicate units, concentrated to their ideational minimum, and their modifying elements. They differ from the three linked, unmodified propositions in a typical Aristotelian syllogism: the premises and the conclusion. The terms in logical propositions repeat in a set order, expressible symbolically as A → B, C → A, and C → B.

37Here is an example:

Given that 1. Humans are mortal. A → B

and given that 2. Socrates is human C → A

it follows that 3. Socrates is mortal. C → B

38The first two propositions share a common term as a premise (human), and the last proposition joins the two unrelated terms in the preceding two premises (mortal, and Socrates). Propositional analysis in cognitive psychology, in contrast, concerns the mental status of any subject-predicate unit, simple or modified.

  • 13 Cognitive load theory (Sweller 1988, 2006) shows that students often fail to solve a problem in ma (...)

39Walter Kintsch (1998: 69-73) describes half a dozen experiments about how test subjects mentally process propositions. One shows that we retrieve a proposition and its modifier differently. For example, if at some point test subjects are primed with a sample sentence, ’Socrates posed moral questions that never failed to annoy his wealthy students,’ and later cued with the word ’questions,’ these subjects would recall words from the core proposition (e.g., ’Socrates’) better than ones from the modifier (e.g., ’students’). In another experiment, when subjects had to remember instructions, ’Doubling the number of propositions from two to four caused an increase in errors from 3% to 52%’ (1998: 70), but doubling the number of words made no difference. Kintsch himself co-authored an experiment that showed that, for every proposition added to a text, reading time increased by 1.5 seconds. These results point to a capacity limit in maintaining in memory any proposition, that is, an utterance converted from natural language to its underlying conceptual meaning. Four terms in two propositions, read in three seconds, cause no trouble, but eight terms in four propositions, read in six seconds, cause cognitive breakdown. These numbers recall Cowan’s two-second phonological memory capacity, 4 ± 2, yet they represent not sounded words but thought. Word-counts are not the right measure for cognitive capacity because readers often model (understand) text in terms of the propositions to which it can be reduced, and the word-count of an easily grasped proposition can well exceed the capacity of working memory. However, if we substitute propositional terms for words in Cowan’s ’magic number,’ the same formula seems to apply to cognitive capacity. Could it be 4 ± 2 propositional terms, if we keep in mind that, for an expert, a term may well point to a substantial schema in long-term memory?13

40Do poetic forms offer corroborating evidence? Early verse forms such as the four-beat, two-part oral-formulaic Anglo-Saxon line link two short phrases by means of alliteration. In 1979 John B. Lord observed that Miller’s ’magical number’ constrains line-length in Anglo-Saxon verse, but the maximum number of lines that form a unified thought in Old English is hard to determine because no stanzaic forms existed then. However, we can measure the number of syllables in verse segments by poets such as Chaucer, after writing became common in England (a form of artificial memory). He used, not alliterating initial word-syllables, but nine- or tensyllable lines bounded by terminal rhymes. If verse forms might be said to define a passage length, Chaucer’s maximum-length verse unit was not the rhyming couplet in the Canterbury Tales but the rhyme-royal stanza found in Troilus and Criseyde: seven pentameter lines, mainly end-stopped, that rhyme ababbcc and have about seventy syllables or 56 words. A contemporary manuscript image shows Chaucer reading Troilus to a court audience, people with some education. He must have assumed that courtiers – who were not scholars – could hold and comprehend, as coherent thoughts, passages much longer than the capacity of working memory. Does Chaucer’s rhyme royal conceivably mark an omega value, the cognitive capacity of a thought?

41Let me convert the words in a popular hymn stanza to propositional form, Isaac Watts’ ’Man Frail and God Eternal’ (beginning ’Our God, our help in ages past’), written in common measure. Its eighth stanza has six different terms (time, stream, bear everything away, sons, dream, and die at dawn) in seven linked propositions.

Time, like an ever-rolling stream,
Bears all his sons away;
They fly, forgotten, as a dream
Dies at the op’ning day.

  • Time is like a stream.
  • A stream bears everything away.
  • Time bears everything away.
  • Time bears away his sons.
  • Time’s sons are like a dream.
  • A dream dies at dawn.
  • Time’s sons die at dawn.

42The fact that the word-count, 21, overflows average working-memory capacity seems to have had no effect on the ease with which ordinary people comprehend this stanza, but this sequence of seven propositional terms exceeds the 4 ± 2 capacity that the experiments of Kintsch and others suggest for cognitive capacity. Is the omega value larger?

43I am guessing not. Most readers will not transform poetic text into logical form in order to comprehend it. They will likely process language as words, images, and episodes in a narrative: that is, they will simultaneously use two or more of the phonological loop, the visuospatial sketchpad, and the episodic buffer in working memory. Remembered images have the same 4±2 constraint (foci) as the phonological loop has for auditory speech (terms); possibly the episodic buffer may have some such limit as well. If so, and if we use the full resources of our working memory to be conscious of a verbalized thought, might it not expand to 12±6 elements in all? Watts’ stanza has eight propositional terms (the subjects and objects time, stream, sons, dream, and dawn, and the verbs be, bear away, and die) and three images (time the river bearing sons away, sons flying forgotten, and a dream lost at dawn).

44Our cognitive capacity could thus be a multiplier of the alpha value. If thought exists partly independently of the language in which it is expressed – and the behaviour of primates indicates that it does − we could understand the meaning of a sentence that is too large to fit, as auditory language, in working memory by activating the non-verbal slave systems of working memory. Cognitive capacity, the phonological loop, the visuospatial sketchpad, the episodic buffer, and the visual reading span might all be defined in terms of a single comparable constraint.

6. Reading Authorial Process

45How can we determine an author’s cognitive capacity? A reading method – not for meaning but for process – cannot be expected to mature quickly because it waits on scientific research of brain function. Yet, I have stumbled over some heuristic techniques. The extent of text that an author can write before repeating himself might well signal his omega value. Shakespeare’s appears to have been about as long as a 14-line sonnet (Lancashire 1999). The output of a blind author, someone who could not store his utterances in writing or by dictating to a machine, might establish a limit: in creating Paradise Lost when blind, Milton reputedly could create and hold 20-30 decasyllabic lines a night before uttering them to an amanuensis. Or, when an author edits a written version of a text, the average size of his interpolations – newly-minted and inserted passages – might represent a cognitive limit.

46Alpha and omega values may not themselves be stylistic markers but they point to what we should be looking for. Although chunk size is universal, the lexical combinations that comprise an author’s chunks will be idiolectally distinctive. Cognitive capacity does not vary from one person to another, but the schemata that authors, as experts, expand to use that capacity and can rapidly shuttle in and out of working memory reflect individual expertise. (A critic must work hard not to impose his own schemas in interpreting an author’s schemas.) Authorship attribution might also conceivably develop measures for how authors express propositional content. We might use the ratio of a proposition’s length in terms and modifiers, to the word-count of the corresponding text into which that proposition is translated, as a stylistic marker. Francis Bacon’s Essays are as renowned for their Tacitean concision as Sir Philip Sidney’s prose is for its Ciceronian amplitude. On average, it seems likely that Sidney will take more words, Bacon fewer, to declare the same proposition or thought.

47If authors sometimes leave traces behind in rare written drafts and audio recordings, once we learn to read a text for authorial processing, we can detect these traces. Examples are everywhere, but I will discuss only two, the spontaneous flow of Shakespeare’s Addition II in The Book of Sir Thomas More (1910), and Virginia Woolf’s calculated recraftings of passages in The Waves (1976, 2006).

The Book of Sir Thomas More

48Hand D in this unpublished play, found over one hundred years ago in British Library MS Harley 7368, is now accepted by most scholars as Shakespeare’s, from about 1594-95. This offers an opportunity to test cybertextuality. G. Blakemore Evans (1974) describes the 147 lines of Addition II by Hand D as ’an authorial first draft, with vague and carelessly used speechprefixes and with deletions and insertions made in the process of composition’ (1974: 1684). Figure 1, below, a facsimile of folio 9r (part of this Addition), shows 52 lines of text, with 446 word-tokens, of which 243 are word-types, for a type-token frequency of 0.41, a fairly high rate of repetition. The secretary hand has some odd features, like the so-called spurred a (hath in line 7).

Figure 1: fol. 9r, British Library Harley MS 7368 (Croft 1973: I, 23).

49Table 1, below, is a transcription of this page. Two numbers precede each chunk, its line number, and then the number of its syllables. It segments Shakespeare’s chunks according to three boundaries, the verse-line end, mid-line punctuation, and the points of error-repair. Shakespeare’s composition took place in chunks that correspond in phrasal length to the alpha value, self-repaired speech errors, and reading span. Pairs of passages, the first pair in italic, the second in boldface, and so forth, show Shakespeare repeating himself: evidence for chunk-segmentation, as well as of feedback from reading what he had just penned. Marked, in small capitals, are several passages where Shakespeare’s composition slowed down considerably, to judge from letter spacing, word-size, and writing out of a word normally abbreviated. These hesitations conceivably mark moments where his cognitive Editor was at work.

50Blank-verse lines, written so as to be 10-11 syllables in length, are conceptual units for playwrights of this period. Shakespeare shaped his words so as to fit them in these 10-11 syllable sequences, but he often broke lines at mid-point: here, in 24 of the 52 lines. These caesural pauses are marked plainly by a comma or a period, and occasionally by a long space, encoded here as <>. Shakespeare does not routinely use full-stops in his writing. For example, lines 18-19, ’why euen yor hurly / cannot proceed but by obedienc what rebell captaine / as mutynes ar incident, by his name / can still the rout,’ have no sentence-ending period or comma after ’obedienc’? Or lines 24-25 go unpunctuated despite their stream of verbs: ’youle put downe straingers / kill them cutt their throts possesse their howses.’ Rather, Shakespeare appears to use a full-stop as a pause while he waited for the next phrasal unit to emerge. The two dozen set-off 5/6-syllable half-lines show a regular constraint at work, suggesting that Shakespeare composed in phrasal chunks.

51The ten errors that Shakespeare makes reveal his mind in the act of emending them as he wrote. Six of these self-repairs exhibit cybertextual feedback and unveil aspects of his language cognition.

521. In line 7, after writing god hath not le (7), and regretting the repetition of both god and lent from line 3, Shakespeare crosses out god and adds he just to the left of it and decides to qualify the verb by only (and so he excises the first syllable of lent).

Table 1. Transcript of fol. 9r, British Library Harley MS 7368

53The two self-repairs occur within a four-word span. The first error interruption occurs in mid-syllable at le. This indicates the short strokes in which the cognitive engine pumps out utterances, as well as shows that the inner Editor is pre-conscious, seeing a problem at a point preceding vocalization.

54One more error occurs because Shakespeare recognizes that he is repeating himself.

  • 14 Scribal changes are not included in the transcription.

552. Another hand, belonging to a professional scribe (and thus irrelevant for our purposes), crosses out everything in lines 17-19 and adds two replacement clauses, ’in in to yor obedienc’ and ’tell me but this’ above the lines.14 Shakespeare’s only deletion here is to cross out warrs (line 18) and write hurly afterwards. He appears to have recognized an unwise repetition of the word from the previous line. When he corrected this is unclear.

56The first occurrence of warrs takes place 12 words before the error halts Shakespeare. This result could have been prompted either by a regressive saccade or by a conceptual trace or by active iconic memory as he was looking at the page on which he was writing.

57Four self-repairs arise from a last-moment change in immediate syntax that does not affect the preceding text.

583. In line 12, Shakespeare deletes the eighth syllable, ar, just after he had written it, evidently because he realized that it would be hard to fill out the line to 10-11 syllables: he then adds the needed three syllables, as you are. The motive for this self-repair is a problem in metrical line length.

594. At the start of line 16, a change in syntax dictates the deletion of that: Shakespeare remakes a subordinate clause modifying feet into an imperative clause. The deletion must have been made immediately.

605. In line 27, Shakespeare deletes sayeng and replaces it by say to introduce an imperative clause. (The addition of alas alas above, an extrametrical four syllables, is harder to understand: the scribe later deleted these words.)

616. At line 34, Shakespeare deletes why you to introduce a seven-syllable subordinate clause rather than immediately give the seven-syllable exclamation, ’why you must needs be straingers,’ that he instead postpones to the next line.

  • 15 There are four other self-corrections. (1) A worry about repeating his three times leads him to cr (...)

62All these self-repairs15 signal cognitive feedback to the author’s own emerging speech. Note how Shakespeare often makes syntactical changes at phrasal boundaries: only then can his Editor perceive how his unconscious mind is tying a phrase to its successor. Another way of putting this is: Shakespeare fought his own unconscious maker at its weakest, which was in forging long units (sentences) from sequences of short ones (phrases).

63In this page, Shakespeare was affected by what he had just uttered on six occasions when he repeats, with variation, word-clusters or semantic groupings that occurred earlier. One (we have already seen) elicited a conscious error check: the repetition of god, lent, and king (3, 7). Caroline Spurgeon found this semantic combination also in Richard II (1930: 265-66). Three other clusters have repetitions following one another closely: the cluster rise, gainst, and god (at lines 10 and 11), the cluster lift and peace (at lines 14 and 15), and the cluster wars, discipline / obedienc, and riot / rout (at lines 17-18 and 18-21). These occur inside spans that fit into working memory. Two others, however, activate only after a long interval, eleven and eight lines later. The first is the cluster obay / obedience, king / rebell, and name (at lines 5-8 and 19-21). The second is the cluster straingers, cutt / knyves, throts / throtes, howses / abode, and hound / doggs (at lines 24-27 and 35-40). Spurgeon finds these clusters in half a dozen plays as ’peculiarly characteristic of Shakespeare’ (1930: 269). Here they look like feedback from persistence of vision, the effect of iconic sensory memory as Shakespeare kept his eyes on the page; or, from time to time, when at a pause, he let them shift upwards in a recursive saccade. All illustrate cybertextual cycles at work.

  • 16 Later, at lines 16 (forgyven), 38 (afoord), and 51-52 (forme and found), this separation disappear (...)
  • 17 The presence of several large spaces separating words, occurring at lines 11, 16, and 45, may be a (...)

64The last evidence of self-monitoring occurs when Shakespeare’s hand slows down, the writing equivalent to hesitation in speech. A combination of traits in letter spacing, word-size, and expansion of a normally abbreviated word identify three passages as written more slowly than the rest. They are (a) lines 1-3, (b) lines 13, 15, and the first half of 16, and (c) line 45. The two telltale orthographic signals of delay in lines 1-3 are the separation of initial fo from the rest of the word in fo rbid (1) and fo r (3), and the whole or partial isolation of initial or medial t from the rest of the word in cer t ainly (2) and t o (3).16 Note also the separation of g from the rest of the word g od (1), and the isolation of r in ce r t ainly (2). Line 13 has both separated initial fo (in fo ule) and the first occurrence, on this page, of expanded your. Shakespeare normally uses abbreviated yor, the sign of a rapid hand at work, as in lines 11, 18, 31, 39, 42, and 51. Lines 13, 15, and 16 (each slowly penned), however, each has expanded your. The third passage, line 45, exhibits three features of slow penmanship: unabbreviated your, the isolation of t in the odd word momtanish, and the separation of the prefix in (in humanyty).17

  • 18 It may be a portmanteau word, `mome+tan+ish’, i.e. tanned or sunburned like a common mome or fool. (...)

65Each delay in writing coincides with a cognitive problem. Lines 1-3 reveal Shakespeare pausing to work out what More’s speech to the mob will be. Delays at lines 13 and 15-16 appear related to Shakespeare’s repair of errors at the end of line 12, and the beginning of line 16. Line 45 is especially interesting. Shakespeare originally intended this line to be the crowd’s (as the speech prefix all shows), but after he penned the prefix he decided that it would be More’s crowning indictment of the mob. The first content word in this insult, ’momtanish,’ is unrecorded elsewhere in English: whatever it means, Shakespeare invented it, and to this day it must elicit an N400 wave on reading.18

Virginia Woolf’s The Waves (1931)

66Woolf’s most experimental novel is streams of consciousness by half a dozen characters, but it was her own unconscious that initiated those streams. Separated by interludes that describe the sea, her chapters reveal stages in a lifetime. The Waves once existed in two holograph manuscript drafts, three typescripts, and a final printed text dating from September 1929 to mid-1931. None of the typescripts has survived, but the first and second holographs have (Woolf 1976: 30). She typed in the afternoon what she wrote in longhand each morning, but usually without changing anything. Leonard Woolf observed that during typing her ’conscious critical intellect was in control and the tension was less’ (1976: 38). One manuscripttypescript pair thus represents each of a version A (1929-30) and B (1930-31), and she used the former as a basis for the latter. Because versions A and B have deletions and marginal additions, the two holographs actually reveal four versions, initial and revised A, and initial and revised B. In each version, Woolf also sometimes rewrote the same passage several times (1976:40). She stored her composed utterances as text in order to work on them and trusted her editor more than her muse.

67One passage in the published novel (2006: 72) concerns a late-night experience by Jinny, one of her six main characters. It appears once in both versions A and B (1976: 190, 505). Version A, in 69 words, describes Jinny walking down a dark street under windows behind which people had undressed and gone to sleep. Then a taxi turns a corner and casts a white light on her. Version B, in 122 words, is written in the first person, from Jinny’s point-of-view, rather than in the third person. Interpolated in materials from version A is Jinny’s meditation of body-parts she feels – her knee, feet, neck, and head – in 45 additional words. Woolf transforms a narration of successive events in version A into a singular, present experience in version B. The final printed version, in 169 words, adds a further passage, found in the middle of the thoughts and experiences of the first two versions, that describes the street in terms of the burning of street lamps, the rush of pedestrians, and some standing policemen. This has about 48 words. Table 2 holds an encoded transcription of A, B, and the final printed text.

68Chunk size (measured by punctuation) increases over the three versions from 1-3 to 1-7 words, and average sentence-length from 4.9 to 6.3 words. The longer in time that the text survives in external memory, on paper, the longer its length grows. Gradually Woolf’s use of external-memory tools relax the constraints of the alpha value in working memory. Each version also becomes progressively longer by the addition of one more thought or experience. A single thought (in 69 words) becomes two (in 139 words) and then three (in 169 words). Woolf rearranges the material of the previous version, deleting some words, expanding others, but she also injects a new passage into the centre of the old.

69The text grows incrementally. Words deleted in the holographs decrease from 14.8 percent to 12.2 percent, and words passed on from A to B, and from A and B to the final print copy, increase from 34.8 percent to 68.9 percent. Storing the text where it is visually accessible normally stabilizes an author’s text, but Woolf’s working habits show that, although she likely reread Version A in starting Version B, she carried forward only chunk-size phrases from it and she recast the sentences entirely. For example, version A uses the past tense and the third-person, but version B the present tense and the first-person. A says ’everybody must havegone to bedwhich was incredible to Jinny,’ whereas B inverts that structure, ’How strange that they should have gone to bed.’ She thus appears to have started over completely, returning to her muse and its long-term-memory stemmata to regenerate the utterance. The image or word-cluster in long-term memory, strengthened by the afternoon typing, would have held dark, moment, street, and light, Jinny and white, and undressed and sleep, eight items occurring 14 times. Woolf’s authorial process from version B to the final copy was different. Some ten sentences from version B recur in the final copy almost verbatim. Much more editing from the page was going on than re-uttering.

Table 2: Three versions of a passage from The Waves

70The three versions show how chunks structure successive revisionings of a passage in The Waves, but they also define Woolf’s omega value. She does not allow repetition to mark her omega limit, but the two interpolations in the last two versions have a like size. They extend to 45 and 48 words in length and amplify two images or ideas in the original, Jinny’s whiteness and the street. In version A, verbal repetitions of dark, darkness, and undressed, three consecutive events (the undressed going to bed, followed by Jinny’s walking, ending with the cab’s turning the corner), and two images (dark and light − the undressed sleeping and the street in dark, Jinny in sudden light) seem manageable in the three slave systems of working memory. The single added thought includes four images in visuospatial working memory (Jinny’s knee, feet, throat, and head), and the repeated words feel or touch, and silk, look to be manageable. Because everything happens in the present, the episodic working memory slave system appears inactive. Last, the 45-word addition in the printed copy presents three visual foci: street lamps burning, rushing pedestrians, and standing policemen.

71In these examples, we observe two authorial processes, Shakespearean flow and Woolfian inflation. Shakespeare creates spontaneously and corrects only in mid-flow. Once he reaches his cognitive capacity, he sometimes repeats, with variations, what he has just uttered. Woolf spontaneously utters a passage in writing and reads it back into her long-term memory by typing it out. She then either regenerates the same thought in a new flow or she edits the typed copy down on paper. In both instances, she typically interpolates a new, spontaneously uttered thought. She inflates her text from within. From my inadequate samples, Shakespeare’s cognitive load appears to be larger, about 70-75 words to Woolf’s 45-50 words.

7. Conclusion

72Cybertextuality draws on cybernetics, computer text analysis, and cognitive psychology to illuminate authors in the texts that survive them. Readers do not constrain an author’s works; the author’s cognitivity does. Without an author’s desire to say something, without personal emotional drives associated by Antonio Damasio with the pre-frontal cortex (1994), there would be no work. The author fashions its propositions, images, and episodes from his unique associational long-term memory and private, prelinguistic thought. When he unselfconsciously formulates the language for these declarations, he faces mental constraints that have only become obvious in the past century. He cannot directly observe the cognitive process of uttering until the very end. The capacity of his working memory limits, to under two seconds, the amount of language he can consciously work with, and then it has to take phonological form as inner speech. If what an author wants to say lies outside his long-honed expertise, it cannot be shuttled in and out of working memory rapidly and will overflow cognitive capacity. What fits within it emerges, pulse-like, pre-phonologically as unspeakable chunks or phrases, sometimes with errors. To exert some control over this opaque, maddening-at-worst, inspiring-at-best creative process, the author relies on cybernetic feedback from his mind’s equally hidden cognitive monitor. It tracks phrases and halts him in mid-stride when it finds an error. His uttering happens, interrupted regularly by hesitations, paralanguage, and self-corrections. Often it is only the feedback of questions by those who are listening that enable the author to define what he intends to say, and to say it.

  • 19 An outliner is a particular type of digital text editor that allows the grouping of text in sectio (...)

73Texts (manuscripts, printed books, digitalia) mark significant victories by both their authors and our species over these limitations. We have unrelentingly developed both cognitive and mechanical technologies consciously so as to gain control of our making. Both these mental and external technologies create feedback mechanisms and give rise to cybertextual cycles. An expert writer pre-builds propositions in long-term memory as maps or schemas, and assembles an utterance by linking two of his memory systems in a collaboration that is cybernetic. Writing, the first external technology, made explicit the contents of a collective long-term memory and offered a visible space that supplements working memory. Then came the computer. Its outliners19 map symbolic concepts to the ampler language in which texts must be expressed. Digital technology, however, goes well beyond storage to create agents that can interrupt, interrogate, and gloss a writer’s work currente calamo. Text-analysis tools, readability software, and spell-and-grammar checkers offer feedback on features of writing.

74If authorial process can be read, if we can detect how the author operates within alpha and omega cognitive limits, the maker is not entirely inaccessible in the text. We need not capitulate to capitalism, which turns a literary creator into an owner, the bearer of copyright. How we are to read authorial process is of course much disputed ground. I can suggest heuristics only at this time. Repeated phrases (isolated and clustered), shifts in the cursiveness of handwriting, in-process authorial corrections, the extent of editorial interpolations, and other stylistic measures (e.g. changes in vocabulary richness) signal mental habits that comprise the cognitive base of a literary work. When Christians search for the Logos in the texts of the Bible, and today when we use, in a revealing metonymy, an author’s name for his collected works, we anthropomorphize an alien neurological entity that we also have within us but of which we are all, nonetheless, largely unconscious. Cybertextuality does not deny the loss of the creator with a photographed face and a pronounceable name but finds in all texts an anonymous entity. We need authorship attribution methods that can analyse more works than are orphaned in copyright limbo.


1 The works of Sigmund Freud make the acceptance of this simple truth much easier. As Karl Lashley says, we can recall the process of writing only in the act of writing (that is, language-making is a procedural memory; see Squire 1987: 152); we have no memory storage system that can save, for later retrieval, knowledge of the steps in which we create an utterance.

2 See Lancashire 2004b. Cybertextuality extends Espen Aarseth’s term ‘cybertext’ (1997), which is first used in 1991 by the ‘speculative fiction’ poet Bruce Boston.

3 I call them alpha and omega because, in operating within their constraints, our minds approach as close to the Old Testament creative God as creatures of a post-Darwinian world can get. `I was in the spirit on the Lord’s day, and heard behind me a great voice, as of a trumpet, Saying, I am Alpha and Omega, the first and the last: and, What thou seest, write in a book[...]’ (Revelation 1:10-11).

4 Mysak specified ten loops, two of them feedforward (ff) and five of them feedback (fb): thought propagation, word formation (ff), thought pattern-word pattern comparison (fb), word production (ff), actual word product-desired word product comparison (fb), word product-thought pattern comparison (fb), internal multiple-loop speech recycling, word product-listener comparison (fb), actual listener reaction-desired listener reaction comparison (fb), and internal and external multiple-loop speech recycling (19). Levelt’s recent theory of lexical access in speech production (1999) uses the same terminology of feedforward and feedback in mapping cognitive processing.

5 A researcher reading a blurred photocopy of an article, or someone listening to a friend’s goodbyes on a cellphone losing power: cybernetics interprets these cases well. The article and the friend’s goodbyes, however badly distorted by noise, are fixed and analyzable. The reader’s and listener’s puzzled reactions - what’s this? what did you say? − are themselves also unambiguous. Everything here can be quantified. We can measure the percentage of loss in visual and acoustic signals and assign a value to them. That value would take into account the redundancy we build into language, unconsciously, to ensure that, despite plenty of interference, we can still understand what is being said. Given that fifty percent of alphabetic English is technically redundant, as Claude Shannon calculated (1948: 14-15; Pierce 1980: 75; cf. Reed and Durlach 1998), a great deal can be obscured without badly damaging the information in an utterance. For example, if a filter removes all sounds above or below 1500 Hz from speech, as A. B. Wood (1955) showed, there is only a 35 percent decrease in intelligibility.

6 See Geschwind (1979), a classical account of how the brain handles language; and Damasio (1994) and Lieberman (2000), for valuable correctives.

7 I hope that someone else, more qualified than myself, has devised or will devise a better formula.

8 Baddeley suggests that the auditory imagery system, not working memory, may be responsible for inner speech that accompanies reading (2004: 49). If so, the distinction may be marginal because their capacities appear to be identical.

9 We fixate more briefly on function words than on content words (Gleason and Ratner 1998: fig. 5.4).

10 I discuss the omega value for this below.

11 XML (Extensible Markup Language) is a set of rules for encoding documents electronically.

12 The Brown Corpus is a carefully compiled selection of current American English, totalling about a million words drawn from a wide variety of sources. It was first compiled by Henry Kucera and W. Nelson Francis at Brown University, Providence, RI in 1963-64.

13 Cognitive load theory (Sweller 1988, 2006) shows that students often fail to solve a problem in mathematics or physics because they use means-ends analysis, which exhausts their cognitive capacity. However, students who use worked amples (which help create schemas in long-term memory) can operate within theircapacity and learn effectively. This expert technique does not increase the numberof chunks but enlarges the scope of the terms in each chunk.

14 Scribal changes are not included in the transcription.

15 There are four other self-corrections. (1) A worry about repeating his three times leads him to cross out the second his in line 8 and overwrite it with an ampersand. Because it does not follow the offending his, Shakespeare could have corrected it in re-reading the finished speech. (2). In line 21, he deletes th and follows immediately with a traytour: the indefinite article does not suggest a specific individual. (This also characterizes Shakespeare’s usage later at lines 23-24 where he writes a rebell.) He catches this error right away. (3) At line 33, Shakespeare deletes the to in spane or portigall, presumably for metrical reasons. (4) At some point, Shakespeare corrects a spelling mistake, the second r to b in the word barbarous in line 36.

16 Later, at lines 16 (forgyven), 38 (afoord), and 51-52 (forme and found), this separation disappears. Normally, initial t in to is unseparated (see lines 5-6, 11, the second half of 16, 24, 27, etc.).

17 The presence of several large spaces separating words, occurring at lines 11, 16, and 45, may be a fourth trait of hesitation.

18 It may be a portmanteau word, `mome+tan+ish’, i.e. tanned or sunburned like a common mome or fool. See OED `tan’ v., and `mome’ n. 2. Wentersdorf (2006) suggests that `momtanish’ is `a contraction of mahometanish’ but if so the word is still a poser because the OED recognizes only the different word-form, `Mahometish.’

19 An outliner is a particular type of digital text editor that allows the grouping of text in sections that are organized in a tree (or hierarchy) of concepts, an outline.

Table des illustrations

Légende Figure 1: fol. 9r, British Library Harley MS 7368 (Croft 1973: I, 23).
Fichier image/jpeg, 74k
Fichier image/jpeg, 213k
Légende Table 1. Transcript of fol. 9r, British Library Harley MS 7368
Fichier image/jpeg, 211k
Légende Table 2: Three versions of a passage from The Waves
Fichier image/jpeg, 256k


Ian Lancashire is Professor of English at the University of Toronto and a member of the Royal Society of Canada. In 2006 he won a prestigious Killam Research Fellowship in English Literature. He is a specialist in Renaissance drama, General Editor of Representative Poetry Online and Lexicons of Early Modern English, founding Director of the Centre for Computing in the Humanities there from 1984 to 1996, Canadian pioneer in the digital humanities, software developer and author of numerous books and articles. His Forgetful Muses: Reading the Author in the Text is forthcoming from the University of Toronto Press.


Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search