Creativity, Poetic Language, and the Computer

A great king packed in an acorn.
–James Dickey, “The Common Grave”

O life,
Dance like a silent tumbleweed!
–IBM 7094-7040 DCS, "Random Stanza Program 3"

The programming of computers to write poetry belongs of course not to the future but to the present, and is indeed very much in the air. “Cybernetic Serendipity,” an exhibition of machine-made art which attracted considerable attention in London in 1968, included poems by the Manchester University Computer, and English translations (made, one assumes, by human beings) of poems electronically produced in Italy. Computer programs in Cambridge, England, and Berkeley, California, have composed haiku of real charm. Edwin Morgan, one of the exhibitors in “Cybernetic Serendipity,” states that his “simulated computer poems” are designed “to take an ironic but not antipathetic look at the relations that will exist between computer creativity and human creativity,” including “deliberate emulation of the so-called blunders or digressions which at times arise (one would say) creatively within a computer context.” My own experiments in computer-generated poetic imagery have shown that such imagery can be startlingly and unpredictably effective, breaching the norms of expression to attain enhanced meaning exactly as does the language of real poets. Since most such experiments involve the use of some sort of randomizing device, they appeal to an interest in the esthetic qualities of the random which is a conspicuous part of the current scene. This in turn is one aspect of a widespread tendency to repudiate form in art and to deny the artist his form-shaping role: hence such phenomena as the found object (which may be a linguistic object) elevated to the status of art-work, the fictional or musical composition whose elements are shuffled to produce a new sequence every time it is read or performed, the movie-ending determined by audience vote, and the like.

It would seem that, at least so far as poetry is concerned, the computer has already proved itself capable of successfully playing the “Turing game” (a test of machine intelligence, devised by the late British mathematician Alan Turing, of which the Scriven test described earlier is an elaboration): images produced by the computer cannot, without extraneous clues or prior knowledge, be readily identified as less vivid, interesting, or original than those produced by human poets. Given such pairs of lines or passages, one of human, one of electronic origin, as (a) “The yellow moon dreamily / tipping buttons of light / down among the leaves” and (b) “In the tinfoil of the moon, / dangerously, / the roses wink”; (a) “Progress is a comfortable disease” and (b) “The horizon has emptied our purpose”; (a) “Snow sparkles like eyesight falling to earth” and (b) “The brooks stroke their ancient bells,” it would be a percipient reader indeed who could unhesitatingly identify the work of the “real” poet in each case. (The first members of the pairs were in fact written by Denise Levertov, E. E. Cummings, and Wallace Stevens, respectively.) The question arises whether, in turning out these and other similarly effective images, a computer may be said to have performed a creative act. As regards the computers of the present, this question can, I think, unhesitatingly be answered in the negative—provided that the answer is based on a full and acceptable definition of creativity in human beings. There then arises a broader and much more interesting question: given the dizzying onward momentum of electronic technology, will the computer ultimately possess those faculties, including the faculty of artistic creation, which have hitherto been thought “uniquely human”?

Confining myself to the sphere of language, I shall propose not so much a definition as a model of the human creative act, derivable in all its essentials from a consideration of how we proceed in playing a simple word-game. Each participant in the game is asked to think of a list of surnames which also fulfill some arbitrarily set condition as words. Let us imagine a version of the game in which the names must also be words in English denoting colors. Once a player has embarked on a search for names belonging to this category, his mental activity will take the following general form. Two or three common names—perhaps Brown and White—will come immediately to mind. When no more suggest themselves, he will resort to some more or less systematic means of thinking up additional examples; he may run over the names of colors, starting with the primary sequence, in which case he will come on Green and Blue, or he may look at his surroundings and find words for the colors he sees, in which case he may come on Black. So it continues: either he is deliberately going through materials in which new names may be found, or he is temporarily at a loss—his thoughts wandering or his mind a blank. Eventually, as at the beginning, there will occur an event which is of particular interest for our purposes: the occurrence of a suitable name without conscious preliminary. To the extent that the name so presented is at once out-of-the-way and apropos, its crossing of the threshold of consciousness will be accompanied by a start of pleasure and affirmation. The player undergoes a “Eureka-experience”—in psychological parlance, an “Aha-Erlebnis.” The sudden appearance on the mental scene of a previously undiscovered good answer may take place when the game has been abandoned, perhaps when the thoughts of the player are busily occupied with other concerns. (In my own case, the name of the British economist R. H. Tawney came back at an odd moment from some long-forgotten reading assignment in a college course; the name Puce has similarly materialized, apparently belonging to some mystery or ghost story read in childhood which I have so far been unable to identify.)

The model of the creative act in language derived from the name game consists of a series of three stages. The first of these I call “summoning”; it includes the player’s decision to take part, plus his understanding of the rules of the game. The summons so formed is directed toward certain areas of association within those mental contents to which memory has either immediate or delayed access (two such areas, in the case of the game as described here). The second stage I call “convergence.” In this stage, the invoked mental contents are made more accessible than usual to examination, and at the same time are brought into closer than usual proximity with each other. The third and final stage I call “recognition.” In this stage, a match is made, such that items from the previously separate areas of the mental contents are perceived in conjunction as having the specified relationship. (Terms in this description having spatial implications, e. g., “convergence,” “area,” “proximity,” “conjunction,” even “perception” and “item,” are used figuratively, without any intention of suggesting a topography or anatomy of the brain. Yet the aspects of mental activity referred to by these terms, however manifested on the neurophysiological level, would seem essential for the playing of the game: the participant must somehow pass from unawareness to awareness with respect to a relationship between verbal items which are already “in” his mind, and for which he deliberately initiates a search.) It will be apparent that the three stages of summoning, convergence, and recognition are characteristic, with appropriate modifications in each case, not only of the playing of word games but of all linguistic expression, and indeed of all thought in which anyone of normal intelligence answers a question or has an idea. The analysis of the process given here makes no claim to novelty. It has always been known that the conscious and volitional part of the mind can receive enlightenment from a seemingly external source, whether that source is explained as natural or supernatural. The term “incubation,” used in psychology to designate the process whereby the mind moves without conscious attention toward the solution of set problems, dates back over forty years, having been used originally by G. Wallas in The Art of Thought (1926). The three stages I have distinguished as components of the human creative act correspond, with certain differences in emphasis, to the three stages of “preparation,” “incubation,” and “illumination” identified, following Helmholtz, by Wallas. (Wallas’s schema included a fourth stage, “verification,” for which an analogue will be found below.) And the culmination of creative activity in an act of “recognition” is figured forth in that most ancient and profound account of the creative process, Genesis 1, where, after summoning the light into being, God sees that it is good. But as I have already intimated, to put a name to what must be taking place at the subliminal levels of the mind is not to understand it. In the words of Sir Peter Medawar, “the ‘spontaneity’ of an idea signifies nothing more than our unawareness of what preceded its irruption into conscious thought.”

With regard to the all-important third step of recognition, we should note well an additional point. The faculty that makes the match in the game cannot be thought of as perceiving, among the totality of the scanned mental contents, only the two matching items. If this were so, we should have to presuppose another faculty to select these two items from the rest, and our model would thereby become vulnerable to the stroke of Ockham’s razor. But if no such additional faculty is presupposed, then the act of recognition must also be one of selection. The making of a match implies, if not the rejection, at least the disregarding of an indefinite number of combinations that do not match. Intimations of this occasionally find their way into consciousness, as when, in trying to remember a name, we keep thinking of and rejecting wrong ones, or when, in playing the game, we become aware of near-misses: pink—but there’s no such name; Pinkham—but that has an extra syllable. Henri Poincare, in his classic account of mathematical creativity (1908), took note of this mysterious process of unconscious selection, speculating as to “the cause that, among the thousand products of our unconscious activity, some are called to pass the threshold, while others remain below.”

The name game as I have described it has obvious affinities with wordplay or punning, and each of the names compiled by the players does in fact provide the materials for a pun. But between playing a game and making even the most trivial of puns there is an all-important difference: a “real” play on words, whether in everyday speech or literary composition, is a manifestation of an expressive impulse which includes both a speaker’s wish to say something and his aspiration to speak in a certain way, to attain to certain qualities in his use of language, and to produce certain effects above and beyond being understood. The search for names has no motivation but itself, the simple willingness to participate in the game and thus to submit to its arbitrary conditions. Speaking and writing, of course, are also “games” insofar as they involve an acceptance of arbitrary conditions, i.e., linguistic or literary rules and conventions, but success in these games is a means to an expressive end, and the expressive end may be such that the rules are repudiated altogether. There is another key difference between word games and wordplay in actual speech: in an actual pun, the areas of the mental contents which provide the matching items are not designated in advance. The speaker disposed to pun is, so to say, on the lookout for a match, but he cannot predict, any more than can his listeners, the directions from which its elements are destined to converge.

Wordplay, especially at the higher levels at which it is more properly called “ambiguity” or richness of meaning, is an important feature of that most creative manifestation of language, the language of poetry. Still more important—”the greatest thing by far,” in Aristotle’s famous characterization, and “the one thing that cannot be learnt from others”—is metaphor. Like puns, metaphors are generated in everyday speech as well as in literary composition, and it will be helpful to approach the metaphor-making process from the direction of the history of the language, in which its continuous importance as a creative activity on the everyday level is readily apparent. The records of the process, available in historical dictionaries such as the unabridged and shorter Oxford, are the thousands of definitions showing how words in all periods have been transferred to new spheres of reference and become established there. We could scarcely express our attitudes toward each other, or our emotions generally, without availing ourselves of this vast accumulated store of extended meanings. In accordance with established metaphorical usage, and for good and sufficient reasons, a person may be called a bitch, cat, dog, fox, goat, skunk, snake, or worm; may be said to bark, bellow, bleat, cluck, coo, hiss, roar, or twitter. Words themselves may be described as acid, astringent, biased, caustic, colorful, dirty, inflated, muddy, ponderous, transparent, or twisted; they may blast, blister, burble, cut, dazzle, flash, gush, scintillate, smear, or thunder. I have deliberately listed only metaphorical meanings which are recorded later than the original meanings of the words in question, meanings which we can, as it were, see being introduced into the language. Since the words put to these metaphorical uses are for the most part familiar, and since the resemblances underlying the metaphors are obvious, it is probable that many of them were “originated” on a number of occasions by a number of different speakers, eventually catching on and becoming established through a process the dynamics of which are little understood.

Both the abundance of this metaphorical small coin and its indispensable character make obvious the inadequacy of those historical accounts of semantic change in which emphasis is placed almost exclusively on the filling of “needs” arising from new cultural developments. Thus in one account we are told that “some words, which otherwise have perished, have been preserved for the sake of a new and definite meaning. Some, in order to meet the constant changes in thought and the progress of knowledge, have had fresh functions assigned to them. New words may be manufactured at any time, but a change in the meaning of an existing word is often a simpler and more convenient way of supplying a need” (Bernard Groom, A Short History of English Words, 1934; reprinted 1965). There is no denying the importance of cultural necessity as the mother of linguistic invention; the need-filling process is responsible for a vast array of metaphorical meanings, among which such comparatively recent examples as advertisement, antenna, bulb, cell, circuit, complex, credit, dial, lobby, network, plug, primary, tape, transmission, and hundreds more could be cited. But there is a motive of no less importance working independently of cultural change—else how explain our continuous inventiveness in metaphorically describing those human traits, emotions and acts which remain constant under the sun? This motive can be nothing other than the aspiration toward “saliency” or heightened significance in the use of language. The speaker who uses a metaphor of this sort without having heard it previously must, we imagine, desire a vividness which the literal expression cannot supply; in attempting to gain it, he substitutes a word conventionally attached to another sphere of reference. As each metaphor is established by repetition (thus becoming itself in a sense “literal”), its vividness tends to die away, though remaining susceptible of recovery in the right context. Ultimately it may become a true “dead metaphor,” in which the original meaning is no longer seen as the basis of the transferred meaning, or has disappeared from the language altogether. The expressive motive for metaphor is naturally at its most powerful when the content of language is emotionally charged; hence the endless proliferation and special saliency of metaphors for money, sex, liquor, drugs, and the forbidden generally; hence the nicknames that replace the established name in an affectionate relationship; hence the private languages of lovers. “Natives of poverty, children of malheur, The gaiety of language is our seigneur”—as in literature, so in life.

The act whereby a metaphor is originated can be analyzed according to the model furnished by the name game, though here again, as with wordplay, we are considering not a mere game but an instance of the actual use of language in all its complexity. The “summoning” of the mental contents again proceeds from an expressive impulse in which the speaker’s intention of saying something is bound up with his stylistic aspirations, including the desire for saliency discussed earlier. The resultant “convergence,” not only of verbal materials directly relevant to the matter in hand, but of other materials linked to them by resemblance, chance association, symbolic identity, phonetic coincidence— all that Coleridge calls the “hooks and eyes” of thought—requires a deployment of psychic energies at the unconscious level whose scope and intricacy cannot be grasped in literal terms and challenge figurative description. To his own question, “How do we think of the words we want to say?” Dwight Bolinger answers, “Our mental scanners seem to sweep over vast networks of words and phrases with deep and invisible connections.” John Livingston Lowes had earlier described “that thronging and shadowy mid-region of consciousness which is the womb of the creative energy in poetry.” According to the psychoanalytically- oriented formulation of Joost Meerloo, “The unconscious knows itself in contact and continual interaction with reality through thousands of threads of communication.” However we conceive of it, this dynamic convergence at the edge of consciousness of ideas and their verbal embodiments contains the materials for the match which is thrust into awareness in the final stage of recognition. In metaphor, however, the match depends, not on an identity of phonetic form between items belonging to two areas of association, but on a relationship of consonance, both conceptual and emotional, between a word and some aspect of an area of experience with which it has not been associated previously. If this relationship can be grasped by those who hear the word in context, the metaphor is capable of entering the public domain. The conceptual suiting of the word to its new subject of reference gives it in effect a new meaning, while the original meaning and associations remain· present, as it were, in suspension. This doubling of meaning combines with the shock resulting from the sheer unexpectedness of the verbal item in context-the new collocation attendant upon any new metaphor—to create the heightened significance or saliency sought. As in wordplay, the stage of recognition in metaphor-making must be considered as also a stage of selection during which, at the instant when speech is formulated, an indefinite number of “candidates” for metaphorical use are disregarded. And here again, the metaphorical match eludes prediction. One may manufacture a metaphor, in the sense of deliberately choosing the fields from which and to which a term is to be transferred, only to find that the same unpredictability holds within a smaller compass: one cannot know in advance from what part of the field the successful candidate for metaphorical use will emerge or what part of the new field it will fit. Our awareness of saliency in language is bound to be after the fact. This is not really surprising; even the results of so lowly a linguistic activity as the name game cannot be known in advance, whether it be the number of names a given player will be able to think of, or the total number of names in existence which fulfill the set conditions. A brief consultation of the Manhattan Telephone Directory recently revealed listings under Pink, Maroon, and Orange, names whose existence I had not previously suspected. Even if a definitive list of names denoting colors could be compiled, it would not necessarily remain so, as there is nothing to prevent new names belonging to the same category from being coined or re-shaped from existent ones: why might not a man named Fuchs change his name to Fuchsia for sake of euphony, thus repeating the process whereby the flower of that color was named for its discoverer? The open-endedness of language at every level is a sign of its vitality.

If the above account of metaphorical innovation is accepted, the actual creation of a metaphor proves to differ from its repetition in early stages of its adoption into the language only in part. The “second speaker” who hears the original match between word and new context must possess or acquire the elements it is composed of as part of his own mental contents, and he must recognize the consonance on which the expressive value of the metaphor depends. The metaphor must then occur to him independently, at a moment which cannot be predicted, as having the same sort of saliency for his purposes as that responsible for its original use. The activity of those whose speech tends to reinforce new means of effective expression exemplifies what Paul Tillich, in The Courage to Be, called “creative participation” in language—”the interdependence of the creative poet or writer and the many who are influenced by him directly or indirectly.” To this we should add the influence, on the everyday level, of the creative speaker.

The achievement of saliency or heightened significance in verbal expression which marks the creative speaker on the everyday level is a fortiori the mark of the gifted poet whose works become part of the cultural tradition he inherits. At its best, the salient image in poetry at once startles and comes home to us, like a returning memory we did not know we had lost. To the practitioner of verbal criticism, such an image, as Jacob Bronowski has said of the scientific fact, appears as not merely an item of descriptive content given form in words, but as “a field—a crisscross of implications.” One of the lines of meaning that cross in the metaphorical substitution of king for oak-tree in Dickey’s “a great king packed in an acorn” is the historical role of the oak itself in religion and myth, the emotions naturally and universally elicited by the tree, wherever found, when it has attained to great age and size. Thus there is the widespread association, among ancient peoples of Western Europe, of the oak with the sky god or lightning god; it is the magna Jovis quercus, the “great oak of Jove,” as also of Zeus; it is the sacred oak of Celtic, Germanic, and Slavic ritual. An echo of this ancient veneration persists in the English language in such expressions, now considered mere descriptive cliches, as “majestic oak,” “stately oak,” “ancestral oaks,” “monarch of the forest,” and the like—and in English poetry in such lines as Spenser’s “builder oake, sole king of forrests all,” Herrick’s “proud Dictator of the State-like wood,” and Cowper’s “lord of the woods, the long surviving oak.” The proverb “Great oaks from little acorns grow” is relevant to the king-acorn image in both wording and content, and we should note also the established metaphorical meaning of the word crown with reference to the leafy head of a great tree (a compound green-crowned occurs in Dickey’s poem in the immediate context of the line in question).

The idea that something of great magnitude and importance is contained in an acorn has a parallel in the figurative expression “in a nutshell”; one is reminded also of folk tales in which nutshells open to pour forth their magically stored contents. Additional strands of significance come together in packed, whose established metaphorical meaning referring to crowds (in “a packed house [theatre],” “a packed subway car”) is here transferred to a single person. To be packed in is not simply to be contained but to be compressed or crowded; with reference to things contained in boxes or articles of luggage, it implies being folded up, and this implication is relevant to the contents of an acorn in that a seedling, as it germinates, both enlarges and unfolds. Finally, being packed signifies readiness for a journey. In Dickey’s poem, the journey is from death to life, to “the sun’s great city,” and it takes place in the final stanza as, with further metaphorical elaboration, “an oak tree breaks / Out and shoves for the moonlight.”

The image of a great king packed in an acorn commands a response on two levels of thought simultaneously. On the one hand, it embodies a truth to which we can give rational assent: the oak is in fact in the acorn in the form of what we are gradually coming to understand as the intertwined double ribbon, fantastic in its sub-molecular intricacy, which bears the coded program for the full-grown tree. But the image also works counter to rational understanding in that it asserts the simultaneous existence of seed and tree, and in that it portrays the tree as a conscious human being, a king suffering in confinement. This tree-king is a figure of mythic stature, a beneficent and awesome presence merging at the unconscious level with the infantile image of the all-powerful adored and dreaded father. In dramatizing the “green-crowded, green-crowned nightmare” within which he struggles, Dickey appeals to ancient and deep emotions; responding imaginatively to the symbolic drama, we identify with the oak in the acorn as we have done earlier with the skeletons of the dead in the common grave, also under the ground but destined never to break free, gazing “over their shoulder” toward the kingdom of being from which they are only a breath away. In that it telescopes time into simultaneity and projects into the realm of vegetative nature a consciousness regarded with sympathy and sacral awe, the image operates at what Sylvano Arieti has described as the “paleologic level of cognition”—that primitive mode of thought, treating similars as identicals and equating the part with the whole, which functions side by side with the abstractive and analytical faculties of the mind throughout life.

In discussing so minor a detail at such length, I hope I have not seemed to be merely retreading the Road to Xanadu. The above is intended neither as an account of what the poet was thinking about during the act of composition, nor as an explanation of the form of the image in terms of specific cultural influences. The relevance of, e.g., the association of the oak with Zeus and Jove, or the line quoted from The Faerie Queene, no more depends on Dickey’s having made a study of classical mythology or having read The Faerie Queene than the relevance of the cited meanings and associations of pack depends on his having read the dictionary. What I have attempted to give is corroborative evidence of a representative nature for the “validity” or accessible form of the image, drawn partly from the cultural traditions common to poet and reader as educated men (including their common language), partly from their shared experience as human beings.

To say that an image has validity in this sense is not to detract in the slightest from its value as an original creation. Though an account of the strands of meaning convergent in it reveals no arcane knowledge, no experience that cannot be vicariously understood, it remains the unique product of a single mind. To make the point simply in terms of verbal collocations (the visible tip of the iceberg which is language in use), the words king, packed, and acorn have not previously been set side by side in a single phrase. Nor would a critic familiar with the work of other well-known poets such as Plath, Lowell, or Berryman be likely to mistake Dickey’s line for one of theirs. We recognize as his the predilection for mythic themes, notably of metempsychosis and metamorphosis; the empathic concern with living nature, especially trees; the stylistic preference for simple language (king where monarch or sovereign might have been used); the flash of immediacy amid a poignant solemnity, here exemplified by the word packed with its everyday associations. The whole conception of the poem—a meditation on the bones “tumbled together” in a common grave amid the endless shape-shifting of a vital principle seen as the “mover of all things” in mist, thistledown, falling leaves, an acorn—bears the hallmark of the poet’s individuality.

Yet the distinctive imaginative preoccupations of a man are not sufficient to account for the poetic product, the fact that language has been cast into memorable form. We have discussed an effective image as a point of convergence, at which many lines of meaning come together and from which they can be traced outwards, but we have not yet taken into account the “meanings” implicit in the phonetic, rhythmic, and grammatical patterns themselves of language, though meanings of this kind are of the essence of poetic expressiveness. An adequate discussion of “a great king packed in an acorn” should include the observation that packed adds significance to the image by virtue not only of its relevant lexical meanings, but of its very presence where it is both grammatically and logically superfluous (“a great king in an acorn” would convey the sense). Packed thus performs the action it denotes, and not only grammatically and logically but metrically as well, since it adds a heavy syllable to make a series of three; its metrical force, in turn, depends partly on its phonetic structure. The achievement of valid form in the language of poetry depends on the ability of the poet to discover “nodal” combinations of words—acorns packed with unfolding meanings—which at once answer to and define his idea of the sort of poem he wishes to write. In place of the two-dimensional match of the name game, what is “recognized” in the act of poetic composition is the coalescence, following upon the convergence of verbal materials in response to the poet’s initiating summons, of many and mutually reinforcing kinds of expressiveness in one and the same series of words. (A similar account, concerned with stylistic effectiveness in language, is presented by Professor Michael Riffaterre in his essay “Criteria for Style Analysis,” 1959; Riffaterre uses the term “convergence” to denote, not an aspect of the creative process as in my account, but its result: the co-presence in small linguistic units of “clusters” of stylistic devices.) It must be true of the poet in particular that his subliminal awareness of language includes a sense of the dramatic potentialities of its sounds and shapes. On the everyday level, this awareness on the part of the speaker manifests itself in the coining of words whose sounds are in some way appropriate to their meanings—the onomatopoeic and otherwise phonically significant vocabulary of a language that springs up in it without recorded authorship.

The act of poetic composition, as Wallace Stevens has said, is “the act of finding What will suffice.” For the poet, as for any artist, the expressive impulse is inseparable from the formative impulse, and this in turn presupposes an interest in a particular medium and what can be done with it. To the extent that the form imposed on language has what I have called “validity,” it is accessible to others: it enters upon a life independent of its creator, fulfilling, in so doing, that desire to transcend mortal limits which Joost Meerloo, in his book Creativity and Eternization, finds central to all art. The salient image in poetry resembles the salient component of the dream as Freud conceived it in that it is “overdetermined,” having many sources in the knowledge and experience of the poet, and in that it exhibits “condensation,” pointing toward a number of its sources at once. The essential difference between poetic and dream imagery is that the ultimate significance of the dream, that sought by the analyst “critic,” is private and personal. The sources of the poetic image are equally personal, of course, but its significance transcends the interests of the individual insofar as it proves self-perpetuating, available to other men in dimensions of time and space exceeding the life of its author.

The response of the reader resembles the creative act of the poet as the repetition of a metaphor by a “second speaker” resembles the origination of a metaphor. If the reader’s response is to be adequate, he must react to many aspects of significance simultaneously in one series of words; as he does this, the image becomes charged with meaning in a process that might be termed “reconvergence.” He is no more aware of these aspects separately than he is aware of dictionary definitions, grammar, level of diction, tone of voice, degrees of stress, and so on, as separate aspects of the language he hears in conversation, though each of these may play a part in determining his instantaneous reactions. He undergoes, in fact, a start or shock analogous to the poet’s own start of recognition as the words that will suffice spring into his consciousness. In this instant of secondary recognition, he repeats imaginatively his own version of the experience to which the poet has given valid form, thus at once confirming and enlarging his sense of personal identity. The deep satisfaction attendant upon the culmination of form in a response adequate to it is what accounts for the difference between the “carrying power” of scientific and poetic discovery noted by Sir Peter Medawar: the fact that, in the latter, “the delight and exaltation … somehow communicate themselves to others. Something travels: we are carried away.” The poetic journey has both its occasion and its goal in human experience; the vehicle of the journey is language having that enhanced significance in which validity or accessible form consists.

Both the creative act of the poet and the responsive act of the reader have a relationship to the poem quite different from that of the critic, whose own activity is that articulation or “unjoining” whereby coalescent strands of significance are taken apart and paraphrased, the simultaneously operative causes of the validity of an image are explained in sequence, and words are found for what, in the creating or reading of poetry, requires no words save the verbal artifact itself and the “stitching and unstitching,” the presentations, discardings, and acceptances which may have preceded the poet’s dismissal of the work as complete. Kenneth Burke said of the language of poetry that “within the word, collapsed into its simultaneous oneness, there is implicit a sequence, a complexity of possible narratives that could be drawn from it,” and it is the task of the critic to draw out those narratives which make themselves manifest to him. But to distinguish the proper activity of the critic from that of both poet and reader per se is not to deny that one and the same person is capable of all three. A man may write a poem, return to it as a reader, and discourse about it as a critic; he may compose poetry and think analytically in rapid alternation. Nor is it to deny to the critic his own creative activity of composing sentences and incorporating them in paragraphs and essays. Finally, it is not to deny the relevance of criticism to the reading of poetry. The critic himself must, and hopefully does, read before he can consider and explain, and his tracing out of relevant lines of significance can deepen the reader’s response to passage or poem. But it remains true that good critics are not necessarily good poets, as good poets are not necessarily good critics, and that the reader who is enlightened by a critical essay does not rehearse the statements made in it at the moment when he re-experiences the poem to which it has enabled him to respond more fully. The work of discursive articulation can take place only after, and never simultaneously with, the discovery of the poetic fact. The image appearing to both poet and reader as a point of convergence is for the critic a point of divergence, a node of meaning already given. The critic can no more predict the image whose power he attempts to explain than the physicist can predict, among all the raindrops in random motion on a window pane, which pairs will converge. Nor, for that matter, can the poet, as Shelley knew: “No man can say, I will write poetry. The greatest poet cannot say it.”

The absence of any necessary connection between the critic’s discursive account of a poem and the poet’s conscious thought-processes during the act of composition is shown by the fact that an image “created” without conscious thought, by a properly programmed electronic machine, can be discussed critically in the same way as an image “created” by a human being. I hope to do this for the computer-generated invocation which serves as the second epigraph of this essay. First, however, I should say something about its experimental background. The aim of Random Stanza Program 3, like that of the other programs in the series, was to produce not so much poems as poetic language, including a high proportion of similes and metaphors. The wording of particular specimens of the output, as opposed to their grammatical form, was to be so far as possible independent of the control of the programmer. To this end, sets of four or five “frames” were devised consisting of function words, such as articles and prepositions, and empty slots. Each slot corresponded to one of a group of vocabularies made up of words fitting certain grammatical, and, in some cases, very broad semantic, specifications (e. g., transitive verb, concrete noun). Once the vocabularies had been stored in the computer’s memory, it was instructed to fill the empty slots in each frame with words selected at random from the vocabularies specified, to print out in stanza form the set of frames thus completed, and to repeat the operation until a given number of stanzas had been compiled. The programs were written in a version of SNOBOL, the linguistically-oriented computer language invented at the Bell Telephone Laboratories in the early 1960’s.

To ensure the uncontrolled character of the imagery generated within the frames, the vocabularies themselves (with a single exception explained below) were selected by random means from the language of poetry in English from Wyatt to the present. To select a fifty-word vocabulary, a traditionally-oriented collection such as An Oxford Anthology of English Poetry was opened at random and a finger placed blindly on the page; the nearest word fitting the specifications was added to the list, up to a total of twenty-five. A second series of twenty-five words was similarly derived from The Young American Poets (1968). In Random Stanza Program 3, a further degree of formal coherence was imposed on the sets of frames with the aid of a vocabulary composed by the programmer, consisting of ten words denoting common poetic subjects—e. g., world, life, city. The stanza was designed to open and close with an apostrophe addressed to one of these. The life-tumbleweed epigraph consists of two frames: (a) “O ___(1)___ ,” (b) “___(2) ___ like a(n) ___ (3) ___ ___(4) ___,” where the four slots are filled by (1) a noun denoting a poetic subject; (2) an intransitive verb, of the sort requiring no additional word to complete its meaning, in the imperative mode; (3) a descriptive, i.e., non-pronominal, adjective; and (4) a concrete noun. For the benefit of such readers as may be curious to see a sample of the output of the program, I append in its entirety the stanza from which the epigraph is quoted, followed by the next three stanzas generated in the series:

O life,
Dance like a silent tumbleweed.
Let the tooth of your sobs
Be famished.
Though all harnesses shine,
Angrily, unalterably, gallantly,
Hallucinate, O life.

O poet,
Let the clapboard of your rage
Be triggered.
Though all stakes probe,
Though all impalas breathe,
Though all moons brighten,
Turbulently,
Decline, O poet.

O autumn,
Let the frog of your brightness
Be ripped.
Though all burrs sink
Though all baskets blur
Impulsively,
Rule, O autumn.

O death,
Teem like a furtive pill,
Close like a furtive clay,
Dwell like an intense clay.
Let the soil of your praise
Be applauded.
Though all tobaccos dream
Though all knees reap,
Shyly, uneasily, steamily, gutturally,
Fall, O death.

The diversity of the poetry from which the vocabularies were selected would, it was hoped, prove an effective guarantee against any controlling bias toward style or subject on the part of the programmer. A certain amount of randomness with respect to the sequences of frames themselves was introduced into some of the programs, though the resultant variations did not affect the output in any way relevant here. No attempt was made to impose prosodic patterns of meter or rhyme, as could be done in expanded and elaborated programs of the same nature.

While I should not wish to claim for the computer’s apostrophe to life, with its simile of the silent tumbleweed, the psychological impact or profundity of Dickey’s king-acorn image, I contend nevertheless that it has poetic interest and effectiveness to a degree that implies the presence of valid form, and hence of a number of convergent lines of meaning. One notes, first, the ancient symbol of the dance for the ongoing physical and vital processes of the cosmos: the dance of the fixed and wandering stars in the cosmology of Plato’s Timaeus, the dances of angels in Christian thought, the round-dance of the disciples, led by Christ, in the apocryphal Gospel of John. (An authoritative discussion of the dance as a cosmic symbol in early thought may be found in Leo Spitzer’s Classical and Christian Ideas of World Harmony.) In English poetry there is the “olde daunce” of sexuality in the exposition of the Wife of Bath, the elegant expatiations on dance as the primal principle of world order in Sir John Davies’ “Orchestra,” the dance of life in the blossoming chestnut tree, where substance and form are one, at the end of Yeats’s “Among School Children.” The word dance implies not only movement in musical measure, but, by virtue of its established metaphorical meaning in such phrases as “dancing flames” and “dancing motes (in a sunbeam),” an irregular bobbing motion like that of a tumbleweed as it rolls onward. It is also relevant to the first component of the word tumbleweed in its now archaic sense “to dance with posturing, balancing … or the like, to execute leaps, somersalts” (Shorter Oxford English Dictionary); the tumbleweed itself both bounds along and progresses head-over- heels. As a spherical object which revolves on its own axis while moving forward, the plant is a natural (though, for obvious reasons of geographical distribution, not an ancient) symbol of the motion of the earth and other celestial bodies. The celestial dance is “silent”; the virtual silence of the tumbleweed as it rolls over a desolate landscape is a part of what makes it strange and haunting to the imagination. “Silent tumbleweed,” unlike “withered tumbleweed,” “rolling tumbleweed,” or even the “tumbling tumbleweed” of the popular song, is both a novel and vivid collocation. The image as a whole is solemn with a touch of homeliness, combining universality of symbolism with local color.

Given the foregoing account of the creative activity of the poet, the question whether, in producing a simile as effective as that of the dancing tumbleweed, the computer has performed a “creative act” in the human sense would seem scarcely worth raising: a machine cannot, at least in the present state of electronic technology, be described as having an autonomous “expressive impulse” governing a summons which causes certain associated materials in its stored memory-contents to converge. But the question still has point as bearing not on the process but the product: on what the computer has accomplished rather than the manner of its accomplishment. Can an effective image appearing in the printout of Random Stanza Program 3 or another program in the series be termed a creative production? If by such a production we mean a new, valid, and unpredictable form imposed on a particular medium, the answer, seemingly, must be yes. I shall return to this point later.

It might of course be objected that the simile of the silent tumbleweed was not unpredictable, and hence lacks one essential feature of a creative production, in that it is merely one of the permutations and combinations of a set of variables determined by the programmer, all of which will inevitably turn up if the program is allowed to run long enough. To this it may be answered that the vocabularies in which the interest and effectiveness of the simile reside were themselves selected at random and were hence “discovered” rather than predicted; the same holds for any particular combination of the words in the vocabularies. If it then be further objected that the contents of the vocabularies are likewise predictable, given the categories and source-texts decided on by the programmer, provided that the random processes by which they are selected are allowed to continue long enough, we have arrived at something like the old fantasy of the thousands of monkeys typing away forever at their thousands of typewriters, fated ultimately to type out all the words that have ever issued, or ever will issue, from human tongue or pen. The concept of predictability then becomes absurd, applicable alike to everything in language from the most ordinary formulas of speech to the complete plays of Shakespeare.

It now seems that we may apply the term “creative” to the results of the computer’s operations while withholding it from the operations themselves. But is the disparity between the human and electronic generative processes really so complete? It is easy to overestimate the degree to which human creativity is volitional and expressive (whether at the conscious or unconscious level), free from all taint of the mechanical or the random. Surely the very contents of the mind, that accumulation of partially verbalized experience from which the elements of any new expression must emerge, are themselves a random input, determined primarily by the culture, era, locality, social class, and family into which a man happens to have been born, and secondarily by the events of his life. Moreover, we must reckon with the arbitrary and rigid resistances of language, as of any medium, to the artist’s attempt to impose on it a form equal to his desire. Especially does this hold true for the language of the poet. “Poetic language,” ancient and modern, has customarily involved a degree of phonetic or rhythmic patterning, or both, above and beyond the patterns of ordinary speech, and the poet working in an inherited tradition is limited by the ways in which the words that occur to him can be made to fit the patterns prescribed. A limited set of consonances, for example, is available in each period to the poet working with end-rhyme; the words that rhymed for Chaucer did not necessarily rhyme for Shakespeare, and vice versa. Rhymes, alliterative links, and metrical values are fortuitous, as are those phonetic identities on which all homonymic wordplay depends. More importantly, the relationship between sound and sense in any language, save for a minority of onomatopoeic or otherwise phonically expressive words, is arbitrary, governed by the accidents of historical sound-change. And each word whose meaning makes it apt for the poet’s purpose carries with it its extraneous baggage, its inadvertent echoes of other meanings and contexts. This, no doubt, was the point of Housman’s observation that the problem in writing poetry is not so much to get the right words in as to keep the wrong words out. The convergencies of sound, sense, and association which constitute valid form in a series of words are in the very nature of the case statistically improbable, and we may justifiably conclude that, a fortiori , the odds against a great poem are well-nigh overwhelming.

Yet these very limitations, however seemingly trivial and mechanical, may provide nourishment for the creative faculty: the mind’s ransacking of its own contents increases the range and depth of self-comprehension, and what is uncovered may be put to felicitous use. If the exigencies of end-rhyme sent Wordsworth to the Hebrides and Emily Dickinson to Van Diemen’s Land, who can doubt that English poetry was thereby enriched? Indeed, it might be argued that the essence of creativity is the ability to capitalize on available resources, to seize, within the welter of inner and outer experience, upon the one word, the one phrase that will serve. What is seized on may be as ordinary and obvious as a given name-as when, perhaps, the Pearl-poet linked the meaning of Marguerite or Margery with the Biblical pearl of great price; as when Chaucer found in the given name of Blanche, Duchess of Lancaster, the white of a piece in chess and of feminine beauty; as when Will Langland saw in his own given name, and used to dramatize a major theme, the same homonymy that Will Shakespeare was to use playfully over two hundred years later; as when John Donne ended his confessional verses addressed to God with the refrain “When thou hast done, thou hast not done.” Or it may be latent in the language: a wedding of expressiveness in sound and sense unexpectedly lending itself to the poet’s purpose—as when Keats discovered in the word forlorn not only the lost remoteness of imagined faery lands (and a rhyme) but a bell tolling him back to the reality of his sole self; as when Whitman heard in the word death the lap and whisper of the life-bringing, death-bringing sea. All such exploitation presupposes the double expressive impulse—personal and formative—that we spoke of earlier; in it we see at work, to quote Coleridge once more, “the genial eye, that saw what it had been seeking, and saw because it sought.” The end of the search justifies the means, even if the means be a dictionary of rhymes or the dictionary itself, where Hart Crane found the word spindrift and greeted it with a howl of triumph. It may even be a printer’s error, such as led W. H. Auden to adopt a previously unintended ports in place of poets in the line “And the ports have names for the sea” (“Journey to Iceland”). We are again reminded that the root sense of the verb invent is “to come upon,” and are humbled to accept Marianne Moore’s salubrious reminder that

Ecstasy affords
the occasion and expediency determines the form.

But what is come upon must also be seen in terms of its expressive potentialities; saliency does not exist until it is recognized, as a forked stick is not a tool until there is someone who knows how it can be made useful. The effective image put together by the computer is not, after all, a creative production in its relation to the machine, for its saliency, its perceptible validity of form, derives not from the machine but from the human eye. The machine itself generates not form but materials, and it is the human reader who, in perceiving form, also confers it, enacting the culminating stage of the creative process. The activity of the computer simulates a part of the process only: that stage, here called convergence, during which stored verbal materials interact in combination and recombination—though with the all-important difference that the contents of the human mind are themselves “programmed” in inconceivably complex, many-leveled systems of association. Like any other machine, the computer serves as an extension of human energies, a marvelous though costly plaything for the manipulation of language, threatening, finally, like the magic bucket in “The Sorcerer’s Apprentice,” to inundate the recognitive power which must operate upon its productions if they are to make the passage from potentiality to realization.

We have so far been dealing with the output of the computer solely in terms of the exceptional case—the occasional image possessing a degree of emotional and imaginative appeal that makes it stand out from the rest. But the poetry generated in any of the Random Stanza programs also has certain stylistic qualities considered as a whole, qualities more or less appealing, depending on one’s tastes. These qualities are dramatic, of the verbal surface; they depend upon the violence with which the norms of collocation in English are continually breached. The computer, as J. R. Pierce has observed, is less inhibited linguistically than a man; read in rapid succession, any sequence of stanzas becomes a circus of semantic and stylistic cacophony, now grotesque, now loony, now indecent, now pathetic; a richly fantastic double-talk teetering precariously on the brink of sense; a lunatic’s inspired flight into poetic high style. These qualities, unlike the profundity vested in the occasional image by chance, are due to the design of the program, which is to say, to the programmer himself. The compilation of vocabularies by random means from poetic texts of the greatest possible diversity does ensure freedom from “authorial” control in the wording of each of the filled frames. But it also opens Pandora’s box, ensuring the constant stylistic jars and reversals, the nonsensical shifts in subject matter, which make for the dramatic qualities described above. In addition, the frames selected for the program affect the qualities of the output in two ways. First, they are loaded with meaning, containing an abnormally high proportion of the words of heaviest semantic content, the nouns, verbs, adjectives, and adverbs; the pronouns, modal verbs, and pronominal adverbs and adjectives are excluded. Second, the invented grammatical frameworks have literary and elevated stylistic qualities in themselves, regardless of the words inserted into them. Thus the structures contained between the opening and closing apostrophes of Random Stanza 3 (pp. 500-01 above) are high-flown and rhetorical, implying imperiousness and vehemence of statement regardless of what the statement is about. Different structures and different methods of vocabulary selection would have produced different results. Even more strictly determined by the nature of the program are those computerized verses, exemplified by the Berkeley and Cambridge haiku, which are put together from vocabularies deliberately designed to produce certain imagistic and tonal effects.

But as regards the output of any program for computer poetry considered as a whole, the role of the programmer goes beyond the authorship of the program itself. Once the machine has been given instructions enabling it to turn out verse of a certain quality and kind, new problems are posed by its virtually limitless energies and the incredible speed of its operations. The output of the Random Stanza Program series may indeed make interesting and amusing reading, but at what point, as we read on, do the interest and amusement pall? Even at the (relatively slow) output rate of two stanzas per second in this series, over seven thousand stanzas can be composed in an hour. How many of them do we want? How many are worth publishing? Questions such as these can be answered in only two ways: by human decision or at random. If the “best” of the works of the computer are to be selected for preservation, the human programmer must choose them according to some human standard of what is best. If a “representative,” unedited sequence of stanzas is to be published, he must determine the length of the sequence. And even if stanzas are to be selected, or a sequence terminated, at random, it is the human programmer who must initiate the generation of random numbers and instruct the machine to proceed in accordance with the result.

In considering the acts of human choice that must follow upon the successful operation of a program for computer poetry, we become aware that, although the three stages of summoning, convergence, and recognition may serve to define the creative activity of the individual, a fourth stage must be included in a full account of the creative process. This fourth stage I call “commitment”; it consists in the cessation of activity on the work now seen as finished (or at least as having attained the best form its author has the power to bestow), and its consignment to the stream of cultural tradition. It is expressed in the ancient topos of the author’s dismissal, the “Go, little book” of the Troilus. Its records are the worksheets that have been left us, showing us the poet making his way to harbor, to the end of the furrow—Keats wrestling through version after version of the casement stanza of “The Eve of St. Agnes,” Yeats discarding line after line, passage after passage, to reach the flawless stanzas of “Sailing to Byzantium.” This stage, of all the four, is the most precarious of attainment, and may in fact lie beyond the compass of the poet himself; it may require an editor, a friend, a Robert Bridges to a Gerard Manley Hopkins. But it remains, nevertheless, the final cause of the poetic endeavor. Even the recluse Emily Dickinson, storing poems away in drawers, could write, “This is my letter to the world.”

Whatever values worthy of preservation the poetical works of the computer may have, they are either determined by the design of the program or brought about by the humanly initiated workings of chance. There is, at least for the present, no other possibility. Given a basic coherence imposed by grammatical frames, the deliberate selection of vocabularies by the programmer will make for consistency of effect, even if the words to be inserted into slots are taken from the vocabularies at random. The less control over vocabulary (and other) input, the more erratic and unpredictable the output. The poetry generated by programs in which relatively little control has been exercised will rise to heights of originality denied to programs on which the hand of the “author” lies heavy, but the human responsibility of identifying that rare occasion when the heights have been reached will correspondingly increase. In any case, the responsibility of terminating the program and deciding what part of its total production is worth the commitment to publish must fall upon human shoulders.

To be genuinely creative in the terms of this discussion, a computer would have to be able, not merely to generate partly predetermined, partly random series of words, but to initiate the writing of poems on subjects of its own choice, to recognize saliency or valid form in verbal expression related to these subjects, to make comparative judgments among alternatives of wording, to cease revising a given poem when a point of maximum effectiveness had in its judgment been reached, and to present its works to the public. Is such a machine conceivable in the long-range future? In order to face this question in its philosophical implications, it is imperative that we concede the practical question, allowing for the possibility that, in time to come, electronic circuitry may attain not only the same density, but the same qualitative complexity, as the neural structure of the human brain. An imaginative leap into the future is called for, an excursus into the sort of science-fantasy that in process of time has so often come true.

The programming of a machine capable of enacting the creative sequence of summoning, convergence, recognition, and commitment would require the installation of multiple and interlinked associative networks, such as must somehow exist in the human mind itself. One such network would be a referential system containing designations for things, acts, events, and ideas, linked in logical, analogical, and symbolic relationships and carrying “charges” of varying intensity corresponding to positive or negative emotional cathexes. To this would be connected a linguistic system, including the vocabulary in toto of English or some other language, complete with phonetic, grammatical, and stylistic information on particular words, as well as a complete grammatical system, a definitive version of the generative grammar of Noam Chomsky or the newer stratificational grammar of Sydney Lamb. Full statistical information would also be needed as to the comparative frequencies of use of words, grammatical constructions, and the parts of speech, both severally and sequentially; this would include data on verbal collocations, cliches, formulas, and proverbs. The grammar would presumably contain information on phonemically significant expressive features such as the patterns of phrasing, word order, intonation, and stress producing assertions, imperatives, exclamations, and questions; this would be supplemented by information on dramatically expressive features such as irony, sarcasm, scepticism, rhetorical emphasis, and the range of emotional tones. In addition, there would be a “lexicon” of the potentialities for dramatic expression of the shapes and sounds of words themselves, both singly and in groups.

Assuming an electronic intelligence so furnished, three additional steps would follow. First, the machine would be given an education, over whatever period of time proved necessary, with emphasis in its curriculum on the English poetic tradition and relevant cultural backgrounds; other literatures, ancient and modern, would be included as deemed appropriate by the programmer. There would be special instruction in metrics, including rhyme, alliteration, assonance, and other patterns of sound, as well as in the figures of speech and of thought. In carrying out the education of the machine, the programmer or programmers would simply be reenacting the roles ordinarily assumed by the parents and teachers of a human being. Second, to ensure individuality in its productions, the machine would be given a set of biases relative to poetic subject matter, style, and form in all its aspects. Insofar as these biases were deliberately imposed by the programmer, they would correspond to the influences exerted by important persons in the life of a human being; insofar as they were random in nature, they would reflect the random factors in inheritance and genetic development from which human individuality springs. Finally, the machine would be incited to compose poetry through the incorporation of servomechanisms whereby the creation of sequences of words having saliency or valid form (defined, perhaps, in some such terms as in this essay), and their combination in larger sequences, would be encouraged, to the point where poems comparable in the relation between length and subject to those studied during the educational period would be produced and submitted to the programmer. The programmer himself might well exercise his own tastes, “rewarding” or “punishing” the machine according as its productions seemed to him more or less original and important, and thus continuing its education; if he chose to do this, he would simply be reenacting the role ordinarily enacted by the editors and critic-friends of the human poet.

Let us concede that this ultimate machine would indeed be capable of writing poetry, or that, if it were not, it could be further programmed to enable it to do so. As we contemplate it in its authorial function (would it be made to print its output on tape or to inscribe it with an intricately automated hand?), the question inevitably arises whether we are not witnessing merely a simulation of the creative functions—a wondrous effigy, empty at the core—as, in a machine programmed to make sounds similar to human expressions of pain when subjected to varying degrees of injury, we would have merely a simulation of pain and not pain itself. If so, the replication, as opposed to the simulation, of mental processes (I owe my understanding of this important distinction to Professors Frederick Crosson and Kenneth Sayre in their essay “Modeling: Simulation and Replication”) has not yet been accomplished. Most would agree, I think, that it is to be looked for from the laboratory, not of the electronics engineer, but of the biochemist.

For the sake of the argument, however, let us concede even this point, and suppose that our ultimate machine does in fact have a mind which it expresses in its poetry. The witty observation of Alan Turing, quoted by Mrs. Sara Turing in her biography of her son, then becomes apropos—that a poem written by a machine would be of interest primarily to other machines. This is not true, of course; such poetry would have a degree of interest for us at least equal to the interest we should take in any manifestation of life on other planets, or in other solar systems. But what, aside from the satisfaction of our intellectual curiosity about it, would it mean to us? Why, once the creation of poetry by the computer had been proved possible, would it be worth our while to cause it to be created?

It is here that we must remind ourselves that the expressive impulse giving rise to the creation of poetry has its source in human experience, and that art achieves its ultimate purpose in helping to define and enlarge the experience of those who respond adequately to it. We may say that both creation and response are predicated on what are, or have been up to now, “uniquely human” capacities, yet fail to realize that what is most distinctively human is not these capacities in themselves, but their presence in a warm-blooded mammalian creature which, like other such creatures, is instinctual, irritable, vulnerable to pain and hunger, capable of fear, anger, sexuality, playfulness, humor, loneliness, and affection. It is uniquely human, not to understand the logical value of the proposition “All men are mortal” and to draw the appropriate consequences for the man Socrates (a machine can easily be programmed to do this much), but to know, and therefore to feel, that we and all those we love must die. Were it not for this and other aspects of human experience, there would be no such thing as poetry, for there would be “no expression, nothing to express.” If indeed at some future time a machine is developed, capable of having experiences of its own and of giving expression to them in art, the value of that art for human beings must be assessed by human beings. Such assessment is the domain of the humanist, and within it he has an authority which all the technological marvels of the future cannot take away.