Skip to main content


Machine Translation and the Poetics of Automation

Avery Slater

Six years after the groundbreaking appearance of The Mathematical Theory of Communication, elaborated by Claude Shannon and with an explanatory essay by Warren Weaver, another Weaver piece appeared in Machine Translation of Languages (1955). This collection of essays extended the theories of linguistic mechanization that Shannon’s work implied. As an electrical engineer trained in formal logic, Shannon’s engineering papers on the manipulation of electrical signals as substitutes for logical processes had begun emerging in the late 1930s. Weaver, the American chair of the Applied Mathematics Panel during World War II, worked to popularize and spread Shannon’s work across a wide range of technical and scientific fields through what was, in many ways, an act of disciplinary translation. Weaver’s role in Mathematical Theory was to elaborate in prose the mathematical stakes of Shannon’s formulations for informational entropy, a new idea that was, itself, a translation of equations drawn from the study of thermodynamics.

Weaver’s essay in this treatise was decisive in the widespread reception of Shannon’s work, transforming the mathematics of entropy (and its technological control) into a translatable module of mathematical structure.1 Weaver’s translation made apparent a set of mathematical tools available for experimentation in disciplines dealing with any form of communicated data, signals, symbolic flow, or calculation. This 1949 collaboration rapidly shifted not only of the status of human language, but also the terrain of translation, as how to translate signals into machines began to appear more fundamental than translations from one language to another. As The Mathematical Theory of Communication was being published, Weaver also released a personal memo to a range of experts in fields that soon would adopt the nascent methodology of information theory. Weaver’s memo speculated on the possibility of employing the new technology of computers to problems beyond translating firing tables for ballistics. His idea was that computers could be used to unravel and clarify the all but inscrutable movements of what to this day remain known as “target languages.”

Only six years later, machine translation had been granted a dedicated research chair at M.I.T., had found a wide range of implementations across the country, had provoked a similar program in the Soviet Union, and had received a high level of defense funding that only continued to increase.2 Owing to the rapidity of the rise, the editors of Machine Translation of Languages felt it appropriate to include an “Historical Introduction” that described the last few years of the project. Weaver’s introduction characterized the papers in the volume as representing a whole new research orientation and episteme with a mythic resonance: “Students of languages and of the structures of languages, the logicians who design computers, the electronic engineers who build and run them…are now engaged in erecting a new Tower of Anti-Babel,” he proclaims. The logic of this tower figure begs the question: what is the new “anti-” Babel “against” in its renewed incarnation? Machine translation’s Anti-Babel, Weaver explains, must “build part of the way back to that mythical situation of simplicity and power when men could communicate freely together, and when this contributed so notably to their effectiveness.”3 Yet as he describes how this new tower is “aimed” in a beneficent direction, Weaver betrays not only the military unconscious behind his metaphor but also a philosophy of language whose enemy was linguistic plurality. Even as it seemed to facilitate the pre-existing multiplicity of earth’s languages, postwar machine translation proceeded from the view that linguistic traits of incommensurability, ambiguity, figurality, idiom, and opacity were so much noise occluding the simplicity and power language could otherwise embody. Drawing on Peter Galison’s theorization of World War II cybernetic technologies as designed around an “ontology of the enemy,” I suggest that machine translation, from its earliest days, is predicated on an ontology of enemy language.4 In this ontology, not only is enemy language the language spoken by a group deemed an enemy. Even further, language itself is attributed subversive, enemy traits in its “resisting” decryption, defying grammatical law, and hiding its intuitive patterns in plain view.

Excited by Weaver’s promise of “simplicity and power” in communication, the U.S. government spent $20 million to fund MT projects at seventeen different institutions between 1956 and 1965. This new ambition to use computation to construct an operable linguistic plurality satisfied political and economic demands to replace the unpredictability of human translation with a process freed of the mess of equivocation, ambiguous terms and undecidable contexts. Within this early machine translation era, an overwhelming ethos of interdisciplinary transformation was afoot, as methodologies from technical, mathematical, behavioral, and linguistic sciences all converged, coordinating previously disparate aims toward concrete projects of Cold War technology. Meanwhile, this era resulted in profound epistemic reconfigurations of the theory of language itself, reterritorializing not the “essence” of language so much as the status of linguistic difference: from socio-cultural field to mathematical, informational quantum. If language seems a different sort of entity to us now in the age of Google Translate, a reconstruction of the technological genealogy for this transformation will show us how this period’s attention to the registration, quantification, and eventual commodification of language recast an ungovernable field of difference into an operable systematicity. From differences to differentials, new theories bent language toward the navigable, the productive, and the predictive.

Much seems to have changed since the early days of unwieldy Cold War computational apparatuses designed to imitate mechanized dictionaries. Machine translation seems no longer in the thrall of rule-following dictionaries governed by simple substitutions, or, as Weaver put it, the “pseudo-mechanical translating device which (like some of the fake mechanical chess-playing machines) had men hidden inside of it, preparing the foreign text for the use of the machine.”5

Machine translation programs of today use complex statistical modelling and (most recently) deep-learning algorithms self-training on online human translated linguistic corpora. Yet while Weaver’s 1955 introductory essay alluded to Wolfgang von Kempelen’s eighteenth-century chess-playing automaton – revenant trope in the techno-imaginary of Western modernity – it also placed this revenant alongside the ancient figure of the Tower of Babel so as to articulate its project of universal translation. The dream connecting Weaver’s Anti-Babel of linguistic simplicity with a chess-playing automaton is one where translation and calculation converge: moving from multiplicity to simplicity, from plurality to power. This was in keeping with larger goals of a majority of Cold War technological innovations to fix or secure desired outcomes, and to do so, paradoxically, by forecasting outcomes: to determine by determining. These tautological forms of modelling-as-control had their apotheosis in postwar game theory, when governing via the politics of containment and intervention made such forecasts into fundamental acts of statecraft. Our latest technologies profit from precisely these forms of statistical mathematics that made game-theoretical risk assessment at a global scale possible.

Indeed, just a few years before the mechanical chess-player would stand in for Weaver’s machine translator, the game-playing automaton famously appeared in Walter Benjamin’s “Theses on the Philosophy of History” (1940). For Benjamin, the automaton was a parable for theology in the modern age, a vestige that stayed out of sight, disguising the animate form hidden within it, so as to impersonate a chess-playing machine. This counterfeit mechanism had a “philosophic counterpart to [its] apparatus.”6 For Benjamin, these were systems of philosophic prognostication that proclaimed history as self-moving (auto-matic), and therefore predictable. Subject neither to aleatory irruption nor to drift, this automaton-theory of history was “to win every time.”7 While Benjamin cast this automaton into doubt, Weaver felt he had helpfully updated it. As the New York Times reported of an early machine translation demonstration in Los Angeles, “when a foreign word for translation is fed into the machine, in the form of an electro-mathematical symbol or a tape or card, the machine will run through its “memory” and … automatically emit a predetermined equivalent – the English word.”8 What Weaver and his colleagues sought was a way of processing the “predetermined” and the “equivalent,” a theory of mathematically computable language such that aleatory shifts in meaning could be contained, allowing a translation that was “right every time.”

Discernible in this early report on machine translation are the cryptographic practices that informed their underlying methods: language was treated as a code, translation was its decryption. Linguistic difference became a challenge to be overcome not by understanding, but by tools of calculation that freed a listener from the labour of understanding. Indeed, this labour struck the information theorists, early computer scientists, and “cyberneticists” of the postwar period as an outmoded form of messy, unaided, prescientific analysis. Norbert Wiener, forger of disciplinary links between artificial intelligence, cryptography, and information theory, saw theories of linguistic and literary translation as mere precursors of more fruitful, cryptographic methods: “the human interest in language seems to be an innate interest in coding and decoding… as nearly specifically human as any interest can be.”9 Wiener’s adoption of the paradigm of “code” to construct his cybernetic theory of language can be traced back to influences such as Weaver himself. In a letter from 1947, Weaver had written to Wiener to interest him in the project of using computers for translation, suggesting to Wiener that “powerful new mechanized methods in cryptography – methods which I believe succeed even when one does not know what language has been coded,” lead one to “naturally wonder” whether “the problem of translation could conceivably be treated as a problem in cryptography.” Weaver continued: “When I look at an article in Russian, I say ‘This is really written in English, but it has been coded in some strange symbols. I will now proceed to decode’.”10

In his later memorandum of 1949, Weaver elaborated:

It is very tempting to say that a book written in Chinese is simply a book written in English which was coded into the “Chinese code.” If we have useful methods for solving almost any cryptographic problem, may it not be that with proper interpretation we already have useful methods for translation?11

In the context of the argument that follows, I will be concerned with the ways in which language is retheorized as code, translation as decryption, and linguistic difference as subvertable through metalanguages (or what we might call “hypo-languages,” symbol-processing systems that are evolved or manufactured to subtend “natural” language). My interest lies in the formation of an epistemē devoted to the computational management of linguistic difference that also leads to the computational “generation” of “natural language” (NLG). Significantly, this endeavor to teach computers how to calculate or “reckon” with language also takes the literary threshold of poetic form as its horizon, the final frontier for its Anti-Babel. Creating the machinic, computational ability to translate was a primordial step in the process of developing machines with the ability to generate language on their own. In the history of computing, translating came before speaking. Technical innovations in engineering machine translation were intimately linked to pursuits of machine cognition, more generally. While translation, for these engineers, lay at the beginning of language and thought, it must not be forgotten that machinic translation was viewed within the larger (mathematical) paradigm of encryption. Perhaps it is this historical kernel of the cryptic or enigmatic signifier that led technicians to their noteworthy choice of heuristic for measuring technological progress in symbol-processing machines: their predecessor enigma, the poem.

Amidst attempts to reverse-engineer linguistic plurality into a homeostasis of machine-assisted monolingualism, the unprepossessing genre of the poem emerges for technicians as a limit-case for their project – consistently cited as constituting the limit for existing technology. The poem, it seems, represented both the highest caliber of linguistic excellence and the limit-point of intelligible communication. The poem was perceived as ancient rival of the new mathematics; its grudging besiegers, reminiscent of the detractors of the modernist avant-garde, quipped “even a [machine] translation of poetry might not be less intelligible than the original.”12 Claude Shannon, whose work in the “Entropy of Printed English” used a pulp detective novel as its linguistic corpus, noted parenthetically that poetry, taken as a sample, led to “somewhat poorer scores.”13

“In language there are only differences without any positive terms,” Saussure has maintained; but will this give us a calculus of differentials? Is there a grid on which lexical infinitesimals can parse not only a memo from a chair of Applied Mathematics, but even Mallarmé’s poetic-noetic flower, “l’absente de tout bouquet”?

Rita Raley’s work on machine translation and the rise of Global English is foundational here, showing us how machine translation’s “functionalist logic requires us to ask how we have come to view the value of language at all: in terms of basic, communicative meaning, or in terms of ambiguity and complexity in the ordinary sense.”14 As Raley makes us aware, the rise of machine translation participates in “a renewed utilitarianism; a renewed appreciation for the basic and easily translatable (the nonfigurative, the non-literary); and a new economics and pragmatics of language and informatic exchange.”15 In the postwar period, poetry’s role in these “pragmatics” of informatic exchange appears deceptively distant. Yet poetry’s new role in the computational age is being written not only by poets but also, surprisingly enough, by computer technicians, mathematicians, and structural linguists. The techno-poetic copula I examine here should be positioned historically as supplementing or extending a process outlined in Foucault’s oeuvre as the dissolution of the concept of the human into the medium of human language, a system of signs, gaining momentum between the end of the nineteenth- and the beginning of the twentieth centuries.16 From the mid-twentieth-century onwards, it should be emphasized that this dissolving of the human has manifested concretely in the way that fictions of expressivity have been incrementally displaced by an ever more powerful digital apparatus, seeming to promise a nonhuman, limitless linguistic generativity to come. Of these fictions of human expressivity more generally, the genre of the poem has served prominently as a canary in the data mines. Poem-generating algorithms, first appearing, by some accounts, in 1959, have since achieved certain consensually acknowledged bars of intelligibility, such as getting published in online journals and passing as “human” poems.17

Yet this AI challenge and these accomplishments have not ceased to proliferate further experimentation. What is being sought in ever more advanced poetic algorithms? An algorithm that can merely generate poems seems insufficient ­– instead, the automatic generation of language is being pursued beyond the point where we can recognize “its” voice, toward the point where we might recognize it as if our own.


Episteme Encoded: Cryptomonolingualism

Postwar Anti-Babel: it began with pessimism about the plurality of languages – as Weaver famously contends, “A most serious problem, for UNESCO and for the constructive and peaceful future of the planet, is the problem of translation.”18 This pessimism had many precursors, for example, as evidenced in the 1930 spring meeting of the International Auxiliary Language Association (IALA) in Geneva, where linguist Otto Jespersen declared it of “utmost importance to civilized mankind” to solve this “really harassing problem” of non-mutual polylingualism. “Linguistic conditions in Europe are desperate,” he lamented: no fewer than 120 languages are spoken (for him, a travesty….).19 “The worst of it,” he claimed, was that while technology made international communication easier than ever, nationalism was becoming stronger than ever, “making each nation feel and maintain, even aggressively maintain, its own value.”20 How could these values be homogenized? How could this plurality, which struck him as excessively costly, be adjudicated? “We have statistics showing the amount paid in customs duties … but no statistics are available to show the fantastic sums and the fantastic length of time spent every year on translations.” Jespersen warned that “the burden of intellectual ‘customs duties’ was undoubtedly heavier than that of material ones.”21

Yet even after the first wave of machine translation, which hoped to do away with those costly human components, the translation laborers, a similar rhetoric endured. Victor Yngve, cosmic ray physicist and MT experimenter, lamented, even as he signed off on the ALPAC report, putting an end to government expenditure on MT, “The world pays a high price for the luxury of many separate languages.” Linguistic difference, that pernicious luxury, might have been eliminated with the computer. The “automatic digital computer could do many tasks besides compute,” Yngve pointed out, including “ease the burdens that the language barriers impose upon us.”22 These luxurious barriers of language were near to circumvention, it seemed, when Weaver first advocated trying translation by computer using “the techniques developed during World War II for the breaking of enemy codes.”23 In the age of Anti-Babel, linguistic plurality was chastened into a sober march toward “the logic of syntax, universal grammar, and semantic factoring… the creation of a possible interlanguage or machinese into which concepts expressed in any given language could be translated.”24 This “machinese” was related to other research on “language invariants” repurposed for computation in code,25 and drew lessons from past technologies (telegraphy) and disciplinary shifts (structural linguistics). 26

Anthony Oettinger, author of the first dissertation in mechanical translation (Russian-English) synthesized binary code with dot-dash notation as two forms of translation: “the transmitting medium requires the telegraphist to translate systematically,” unleashing “electric impulses traveling along a wire” and translated “as holes in a paper tape” that are “used to control a printing device which reproduces the information… In the course of this process the same information has been translated several times.”27 Translation and transcription became synonymous through the combined powers of many, recently emerged apparatuses for quantifying language: Bell Labs catalogs of word frequency, Zipf’s laws governing linguistic frequency in a corpus, Yule’s mathematics of style. “The Epistemic is not a general reason; it is a complex relationship of successive displacement in time,” explains Foucault, describing how an emphasis on “discontinuity” guides his scholarly analytic. Yet, while Foucault rejects the possibility of any “sovereign” optic for making sense of epistemic transformations in history,  he does not neglect to highlight a “new function of language with respect to objects” in the twentieth century, a linguistic project that no longer seeks to penetrate, inquire into, or disclose objects “but to extend a surface of transcription where the form, the number, the size and the disposition of elements can be translated in a univocal manner.”28 Even as Foucault emphasizes the discontinuity of epistemic paradigms within history, we may infer his unspoken awareness that the counterpoint to his own view lies in the growth of a computational, scientific “sovereignty” that makes transcription and translation continuous, seeking to generate an extensible field (of commodities, data, persons) that will be amenable to certain forms of “processing.”

Grown from the logic of enlightenment transparency, the postwar paradigm of code “extends the surface of transcription” in every direction at once, seeming to take natural language as one of its primordial adversaries. Speaking of the interlocking fates of information technologies and the concomitant rise of Global English, Raley has urged us to see that “outcomes and futures of English, even in its current operative incarnation…speak about its becoming-code, its functioning as a code.” Raley concludes, with respect to discourses around English and Englishes, that “neither the narrative of imposition nor the narrative of radical multiplicity and fragmentation can stand”; rather, she challenges us to see the ways in which English is becoming “a basic neutral code operative and operable as a virus, insinuating itself into various networks, with the hosts accepting, not rejecting, the transmission. This is code-switching of a different order.”29

Pursuant to this, we can ask, what does it mean to track a logic of encryption, not only beneath the logic of the catalogued and quantified universe described by Foucault, but also beneath the appearance of a technologically instantiated “Global English”? To see this paradox of encrypted ubiquity, we may take both sides of Raley’s figure of “virus” – something designed both to hack and to violently replicate (machine-translate?) the infrastructures of its host. In the realm of the computer, as Friedrich Kittler’s analyses bring out, a new form of symbolic control emerges through the partitioning of operating system and software, of central processing unit and machine language: here, software “gain[s] in user-friendliness as it more closely approximates the cryptological ideal of the one-way function.” Kittler argues that this trend “cannot be adequately explained either through technical advances or through the formalities of a theory of models;” instead, “like all cryptology,” this encoding that subtends is itself motivated and “has strategic functions.”30

Such hard-wired paradigms of control that resist being read are what lead Alan Liu to ask: “What are the aesthetics of encoded or structured discourse or, as I will term it, of postindustrial dematerialization? And: How is it possible for writers or artists to create in such a medium?”31 Radical and experimental aesthetic practice at the turn of the twentieth century, such as in the case of the Russian futurists and Formalists, considered that the truest function of art might paradoxically be realized by a “device,” a device that defamiliarized the world, stripping psyches of their everyday automatisms: breaking “the glass armor of the familiar,” in Viktor Šklovskii’s words.32 Peter Steiner sees the the Russian futurists’ “mechanistic” tropologies as related in part to the machine as cult object representing the passage to modernity.33 French-Polish modernist poet Guillaume Apollinaire, having elsewhere enjoined poets to reclaim their mytho-genetic powers from machines, wrote in Cubist Painters (1913): “Above all, artists are men who wish to become inhuman. They painstakingly search for traces of inhumanity, traces which are nowhere encountered in nature.”34 This is a complex passage, insinuating a familiarity with Hegel and Marx’s conceptions of “second nature” – a world thoroughly worked over by human technē. Apollinaire continues:

Such traces are the truth, and beyond them we know no reality. But reality can never be discovered once and for all. Truth will always be new. Otherwise it would just be a sorrier system than nature. If that were the case, poor truth, ever further away, ever less clear and less real, would reduce painting to a plastic script intended simply to allow people of the same race to communicate. Nowadays we could soon find a machine able mindlessly to copy such signs.35

It would soon come to seem, however, that only the machine – or machinic forms of practice – could guarantee the renewal of other “sorrier systems.” As Surrealism before and after the Second World War fetishized the “automatic working of cognition,” unfettered by rationality, other projects of chance-driven and arbitrary-constraint-driven poetry proliferated in the postwar period: the Stuttgart aleatorists, American cut-up art, Oulipo, Fluxus, and countless others. In “Linguistics and Communication Theory” in 1961, Roman Jakobson – having seen the arc from Russian Formalism to information theory to cybernetic structuralism, wrote of a “spectacular tie between linguistics, in particular the study of poetic language… and the mathematical analysis of stochastic processes.”36 Jakobson regarded Russian Formalist critic Boris Tomaševskij, among others, as “expert both in mathematics and in philology.” This disciplinary skill-set allowed Tomaševskij to use “Markov chains for the statistical investigation of verse; these data, supplemented by a linguistic analysis of the verse structure,” offering what Jakobson calls a “calculus of [poetry’s] conditional possibilities and of the tensions between anticipation and unexpectedness as the measurable rhythmical values, and the computation of these tensions.”37

Before the poem could come to represent a limit for machines, it had to “descend” in some way to the science of calculation. This scientization of poetics was a profoundly modernist project – in the second sense of modernism outlined by Susan Stanford Friedman in her “Definitional Excursions,” the positivistic, rationalist, totalizing paradigm that sought to unify disparate structures into a metalanguage of intelligibility, the “modern” world.38

If such structuralism gave us, famously, the death of the author, has it also given us the death of the translator? How do the related projects of structural linguistics, mathematical language formalization, and machine translation map onto ideas of the “author as producer” (to cite Benjamin) in the postwar half-century? What transformation is underway once the work of art in the age of its mechanical (re)producibility is replaced, as a problematic, with the work of art in the age of its computational generativity? Jim Carpenter, author of the algorithm Erica T. Carpenter that generated counterfeit poems which more or less successfully passed for the work of well-known, contemporary poets in 2008, wrote enthusiastically, a few years earlier, of “The Poetry Machine” to come that

will have nudged computed poetry out of mere novelty and passing diversion. It will have composed texts worth reading. And most importantly, it will have obsolesced the Author and rendered Him irrelevant. The Author will not just be dead, as Barthes declared, but unnecessary.39

What does this enthusiasm to render the author unnecessary reveal about the present predicament of human and social articulation?  This gleeful ethos of the obsolete human emerges early on in the social and artistic world surrounding the computational “revolution.” Even when the tools to hand were not precisely mechanical, poets were finding new possibilities in experimenting with the notion of algorithm. Brion Gysin, in an essay espousing the “cut-up” technique (developed with William Bourroughs) in a 1964 issue of the Evergreen Review, exclaimed “Who told poets they were supposed to think? … Poets have no words ‘of their very own.’ Writers don’t own their words. Since when do words belong to anybody. ‘Your very own words,’ indeed! And who are you?”40

While these forms of deconstructed authorialism connect to the much-needed critiques of Enlightenment subjectivity instituted by Lacan, Barthes, Foucault and others, it is worth considering the extent to which the postwar avant-garde imagination was influenced, in its practices, by the promise of the computer as language-generator. Theorists and engineers of the computer, on the other hand, seem quite enamored of the chance to revive the viability of Enlightenment rationality, albeit outsourced in their machines.

“In a certain sense, all communication systems terminate in machines, but the ordinary language systems terminate in the special sort of machine known as a human being,” writes Norbert Wiener in 1954. Wiener was already by that time a widely read public intellectual and prophet of a new biomechanical order based on the mathematics of communication. One level defining this human-machine’s system of communication was the “semantic,” a fraught and insufficient terrain where “difficulties of translating” result from “imperfect correspondence between the meanings of words”; Wiener reduces a semantic phenomenon to one that “restricts the flow of information from one into the other.”41 The popularity of Wiener’s cybernetic paradigm lay in Wiener’s ability to insinuate a unified field theory applicable to any and all forms of “message”:

Besides the electrical engineering theory of the transmission of messages, there is a larger field which includes not only the study of language but the study of messages as a means of controlling machinery and society, the development of computing machines and other such automata… a tentative new theory of scientific method. This larger theory of messages is probabilistic theory.42

Wiener sought to displace human communication onto ever larger fields of communicative transitivity. In the cybernetic ontology of message and receiver, the notion of complex yet computable communicative transactions – between organism and environment – structured every relation from the cellular to the social, from the broadcast news all the way down to the neural apparatus. “Cybernetics thus stages for us a nonmodern ontology in which people and things are not so different after all,” Andrew Pickering later infers in his laudatory reading of this scientific movement.43 Pickering contextualizes cybernetics as designed beside and against the (humanistic) academy’s postwar linguistic turn, depicting cybernetics as heroically struggling to emerge “in a world where epistemology was the thing, and ontology talk was verboten.”44 While Pickering believes that “the history of cybernetics might help us imagine a future different from the grim visions of today,” it is important to note that Pickering’s optimism comes from his alignment with cybernetics’ generalized attempts to delete representational detours and get to the things themselves: in other words, cybernetics’s information-theoretical project to reconceive language as operable code, as translation as a form of encryption. In this cybernetic view of the organism as information processor, the strength of “biological computing was that it was immediately performative, involving no detours through the space of representation.”45 What stands most clearly as obstacle in the path toward an ontology engineered by cybernetics is the problem of language as something other than computable.

As John von Neumann, architect of the modern computer, game theory, and key advances in thermonuclear warfare explained in the Silliman Lectures of 1956, the “language” of the nervous system “may well correspond to a short code … when we talk mathematics, we may be discussing a secondary language, built on the primary language truly used by the central nervous system.”46 Wiener also classed the neural events of embodied awareness as part of a much larger informational system, open to operationalization in similar ways: “When I control the actions of another person, I communicate a message to him… the technique of communication does not differ from that of a message of fact.” And further: “When I give an order to a machine, the situation is not essentially different from that which arises when I give an order to a person…. I am aware of the order that has gone out and of the signal of compliance that has come back.”47

As Italo Calvino noted in 1967, “Electronic brains…provid[e] us with a convincing theoretical model for the most complex processes of our memory… our imagination, our conscience.” The speech, titled “Cybernetics and Ghosts,” marked the extent to which cybernetic ideology had permeated beyond the laboratory and into the cultural sphere. “Shannon, Weiner, von Neumann, and Turing have radically altered our image of our mental processes,”  Calvino writes, so that rather than “impalpable psychological states and shadowy landscapes of the soul … we now feel the rapid passage of signals on the intricate circuits that connect the relays, the diodes, the transistors with which our skulls are crammed.”48 In this characterization of the predicament of the artist as cyborg in the early years of the computational revolution, we might hear an echo of what Benjamin explained was to be the fate of the “second technology,” a shift away from the first forms of technē in which “technology served as an instrument of ritual.” Benjamin explained that “the origin of the second technology lies at the point where, by an unconscious ruse, human beings first begin to distance themselves from nature. It lies, in other words, in play.” 49

“Society can only be understood through a study of the messages and the communication facilities which belong to it,” writes Wiener, sounding uncannily like Foucault, yet crucially omitting any critique of power such as motivates the latter’s account. Wiener prophesies that “messages between man and machines, between machines and man, and between machine and machine, are destined to play an ever-increasing part” in the make-up of everyday life.50 Not merely for aesthetic reasons, then, do we find, as Calvino claims, that “everything in literature and culture that today – after von Neumann – we see as a combinatorial mathematical game.”51 After Benjamin, furthermore, we find that “the primary social function of art today is to rehearse that interplay” between persons and things, subjects and systems.52 “We aim to rewrite this Show,” promises Gysin, “and there is no part in it for Hope. Cut-Ups are Machine Age knife-magic…. Cut through what you are reading …do cut-ups and fold-ins until we can deliver the Reality Machine in commercially reasonable quantities.”53 The rhetoric of commerce and “Reality Machines” was no flippant hyperbole, but remains rather a real meditation on a present state of affairs. Only toward the end of the 1960s did the economic viability of MT projects built around rule-based grammars become targets of serious skepticism. Crucially, it was the economic rather than the technological rubric of possibility under consideration. With respect to whether or not one could eventually get a computer to translate well, it had started to seem clear that human translators could continue to perform this act more cheaply than computers, given what had been achieved computationally. Even after State Department funding for such projects dried up in the late 1960s, the quest to achieve neural parity through computational logic continued. As John McCarthy, head of Stanford University’s Artificial Intelligence Laboratory proclaimed in 1973, convinced of the “programmability” of mind by logic-based means: “The only reason we have not yet succeeded in simulating every aspect of the real world is that we have been lacking a sufficiently powerful logical calculus. I am currently working on that problem.”54)

The alleged “problem” to which McCarthy referred lay in finding or inventing a uniform, infinitely extensible language for perceptible, cognizable reality: an existential monolingualism. The problem was posed not by the power of nature, but by the language of nature – that is, the language of nature conceived as code.55 Here, cryptological convictions beneath this monolinguistic ideology reappeared: “Nature offers resistance to decoding,” Wiener admitted, “but it does not show ingenuity in finding new and undecipherable methods for jamming our communication with the outer world.”56 The tendency of the nonhuman world toward what we might call a perverse polylingualism represented for Wiener the “passive resistance of nature.” This nature resisted, but not ingeniously. Mechanically? Nature seemed almost an automaton… but not a cyborg. Nature seemed to originate in technology’s first age, an age that, for information theorists, was characterized by wasteful transfers of data, ruinous translation between organism and environment where, Wiener continued, an “act of translation [is] capable of dissipating information… dissipated but not gained…the cybernetic form of the second law of thermodynamics.”57 When one subjects language as such to the rules of entropy (as Claude Shannon did for vocal sound and writing systems), one misses the fact that, whenever linguistic translation occurs – whether it’s judged to be inaccurate or attains the upper limits of fidelity – in all cases, something else comes into being. Literary translation sutures, only momentarily, two systems of ordering the world (languages); any claims to commensurability are tempered by a total rejection of assimilation – the data sets cannot be merged. For this reason, its helpless events of “impurity” produce new meaning, not no meaning.

The problem of “translation” described by its machine-theorists is in fact the problem of a philosophy of transcription and transitivity trying to pass itself off as translation. In the case of MT, there are always at least three “codes” – two linguistic, and one machinic. Thus, while in linguistic translation, one language does something “beside” another, in transcription, one language does something to another. The machinic language “renders” the source and the target; but the transfer is not without residue. Some residue of technical parametrization is left behind. “A mechanical translator, like the sorcerer’s apprentice, is unable to desist,” early MT theorists Richens and Booth admitted with chagrin. “It will continue to translate even when not required, as for example, when it encounters proper names… the readers must be prepared for Tours to come out as turn/tower (plural) and for Mr. Kondo to appear as Mr. near-wistaria.”58 Here the Tours of the proper name that cannot stay a name, but can only be rendered into abstract concepts – towers, turns – might remind us of Babel itself, both city name and words-within-a-name: Ba-bel, or “city of god,” as Derrida pointed out.59 How can the machine be programmed to refrain from rendering such a unit? How will the machine know when the Tower is high enough?

Michael Cronin points out that in Breughel’s painting of the Tower of Babel, technological implements are everywhere. “Arguably, the true common language here is technology itself, as the workers make use of the assorted tools to build the tower to the heavens. In the biblical account in Genesis, language is inseparable from technical potential, from the power to create, shape and transform.”60 We might set this perspective in contrast with Benjamin’s take in 1916, when, in “On Language as Such and on the Language of Man,” he asserted that God did not subject humanity to language, nor to its pluralities, rather, “in man God set language, which had served Him as a medium of creation, free.” Moreover, Benjamin advised us to remember that “the name is no closer to the word than knowledge is to creation.”61

Still, the midcentury pioneers of linguistic formalization believed they had found, if not a language, then a meta-language for creation. Jakobson, quoting structural linguist Leonard Bloomfield’s assertion regarding the continuum between number and word – “mathematics is merely the best that language can do” – updated Bloomfield’s statement by noting the wholesale uptake in linguistics circles of “mathematical logic, in particular, the theory of recursive functions and automata…the theory of communication and probabilistic models.”62 In a paper given in 1960 at a conference on the Structure of Language and Its Mathematical Aspects, Jakobson praised the discovery of a “spectacular tie between … the study of poetic language, on the one hand, and the mathematical analysis of stochastic processes on the other.”63 In Jakobson’s understanding of this “spectacular tie,” “the message is first a stochastic process,” to the listener in which “the probabilistic aspect of speech finds conspicuous expression.”64 In this new view of language, whose “overall code includes a set of subcodes,” and “rules of transformation of the optimal, explicit kernel code,”65 Jakobson expounds that “metaphoric creations … are not deviations but regular processes of certain stylistic variants … subcodes of an overall code.”66 We should not be surprised here to find Jakobson leaning toward the encryption paradigm, in which literary audience, translator, and decrypter converge and become indistinguishable, in a process where “the most detached and external onlooker acts as a cryptanalyst, who is a recipient of messages without being their addressee.”67

Solomon Marcus, in “Linguistics as a Pilot Science” (1963) explained that linguistics now overlapped with “symbolic logic, theory of programming languages, theory of numbers, combinatorial geometry, theory of formal systems, automatic documentation, genetics, code theory, sociology, psychoanalysis, anthropology, musicology, folklore, filmology, and theory of literature.”68 Expressing a rather circular foundational argument, Marcus insisted that “Linguistics cannot limit its object and interest… to natural languages” but must consider computer codes as well, “their structure being very closely related.” Natural language semantics, he went on to note, were then being studied through “mappings from natural languages onto logical languages.”69 Rather than furnish insights about human language, would such human-to-logical linguistic mappings only intensify the stakes of the wager that logical procedures were supposed to answer?70

Emily Apter has criticized a similar ideology in the realm of digital fluidity:

it becomes clear that digital code holds out the prospect, at least, of translating everything into everything else. A kind of universal cipher, or default language of information, digital code will potentially function like a catalytic converter, translating beyond the interlingual and among orders of bios and genus, liquid and solid, music and architecture, natural language and artificial intelligence, language and genes, nature and data, information and capital.71

The fluidity and complexity of “natural language” systems was greeted alternately with scorn and envy by the metalanguage devised to both describe it and master it. For instance, structural linguist Zellig Harris noted that while “the formal analysis of language” discovered “the same kinds of relations and combinations which are devised in logic and mathematics,” languages also exhibited “more complicated types of combination than people have invented for logic.”72

The theorists and technicians pressed on, hoping to tame this complexity and to stake out a field of unity amidst so much plurality. The “cryptographic-translation idea,” as Weaver framed it, was concerned not only with the “statistical semantic character of languages,” but, as mentioned above, was expected to also “make deep use of language invariants.”73 In a strange, Kafkaesque inversion of Babel, he proposed the following:

Think, by analogy, of individuals living in a series of tall closed towers, all erected over a common foundation….  they shout back and forth, each from his own closed tower…. But, when an individual goes down his tower, he finds himself in a great open basement, common to all the towers… Perhaps the way is to descend, from each language, down to the common base of human communication – the real but as yet undiscovered universal language.74

Where Weaver saw a common subterranean passage beneath Babelian towers, Kittler saw a “postmodern Tower of Babel” erected by computation. Here “programming languages have eroded the monopoly of ordinary language and grown into a new hierarchy of their own.” Kittler wanted us to attend to the ways in which “modern media technologies in general have been explicitly contrived to evade perception.” The result was an occasion of unknowing. As Kittler put it, “We simply do not know what our writing does.”75

But is there an art that arises in such a scenario? Of the many poetic experiments that have emerged to mediate and confront the problem of linguistic generativity, I will focus on two that are now regarded as canonical examples in the so-called “prehistory of digital poetry,” but whose original intentions or successes were far from clear to their designers. The first is the work of Theo Lutz, producer of the first “stochastic poems.” The second is “Poetry Vending Machine,” by Hans Magnus Enzenberger, a treatise written in the 1970s in a spirit of “ennui” against radical politics. The work remained unpublished until 2000, at which time a municipal council offered to construct the machine outlined as a thought experiment in the treatise. Both these computational, language-generating projects operate within and on the concept of literature and the “poem.” But they leave us with more questions than they answer, especially when the literary tropes they would seem to transcend are read back against them.

As the heirs of machine translation research, NLG continues to this day to pursue the limit of the poem. In a research article from 2015 we read that “A poem is generated by filling the grid, one element at a time. Each final word is selected from its set of candidate words” through “a step-wise pruning of the whole lexicon by the reasoner, based on various constraints.”76 It would seem that little has changed since the early days of mechanical substitution in translation – though the programs are more elegant, the machines faster, the mathematics more complex… However, that which emerges from the newer experiments is akin to machine translation’s legendary “The vodka is good, but the meat is lousy.” [The alleged Soviet machine translation of the English phrase: The spirit is willing, but the flesh is weak.]. If our machines are now, or have always only been, capable of a Dada aesthetic, are we to ascribe to them the revolutionary motives Benjamin attributed to Dada’s originary proponents, whose “poems [were] ‘word-salad’” made from “every imaginable kind of linguistic refuse…a ruthless annihilation of the aura in every object they produced”?77 The key to the poetry-generating neo-Dada machine is what Benjamin pointed out about Dada 1.0: every object made was branded “a reproduction through the very means of its production.”78 The question is what is being reproduced – the fact of language or the fact of computation?



[4a/b, 4a+a1/b, 4a+a1+a2/b, 4a+a1+a2+a3/b …] This is how Goethe wrote Faust.79

–Andrei Bely, 1909

I like the feeling the everlasting feeling of sentences as they diagram themselves.80

–Gertrude Stein, “Poetry and Grammar”

Les quantités du langage et leurs rapports son régulièrement exprimable dans leur nature fondamentale, par des formules mathématiques.81

–Ferdinand de Saussure

Italo Calvino’s 1967 essay presents an apt framing for these permutations and final questions concerning the imminent “substitution” of the poet/author/translator by the machine:

I am not now thinking of a machine capable merely of ‘assembly-line’ literary production, which would already be mechanical in itself. I am thinking of a writing machine that would bring to the page all those things that we are accustomed to consider as the most jealously guarded attributes of our psychological life, of our daily experience, our unpredictable changes of mood and inner elations, despairs and moments of illumination. What are these if not so many linguistic ‘fields,’ for which we might well succeed in establishing the vocabulary, grammar, syntax, and properties of permutation?82

Joseph Weizenbaum, the computer scientist who invented ELIZA (the first computer program arguably to have passed the Turing Test, by imitating a Rogerian psychoanalyst), questioned this infinitely extensible logic of computability, explaining: “There can be no total understanding and no absolutely reliable test of understanding.” “It is precisely barriers of this kind that artists, especially poets, struggle against.”83 According to Weizenbaum, understanding cannot be defined merely by verifying whether a message recipient possesses an eidolon parallel to that of the sender, “for by that criterion no human understands any other human.”84 How, after all, do we measure our own understanding of each other? For Paul de Man, this predicament of unknowability within language and signification is precisely what translation reveals, namely, “that this alienation is at its strongest in our relation to our own original language, that the original language within which we are engaged is disarticulated in a way which imposes upon us a particular alienation, a particular suffering.”85

Another, monumental experiment in the history of avant-garde automatic poetry is German poet and translator Hans Magnus Enzensberger’s Invitation to a Poemat. This short manifesto, written in a grimly farcical mood, imagines the construction of a machine-poet that Enzensberger did not intend ever to be constructed. In 1974, Enzensberger is already taking a retrospective view of previous computational text-machines, “well-educated cretins,” as he terms them.86 Enzensberger insists that real success can only come from a a “three-dimensional poem,” a textual artifact whose coherence goes beyond simple sentences. Critiquing grammar-programmed and ultimately boring text-producing algorithms, he points to the example of metaphor as lying already beyond grammaticality: “The House” + “sleeps” can be grammatical and poetic, while “The House” + “spoon-feeds” is grammatical yet hardly cogent, it is merely “anomalous.” Enzensberger states that while a rule-following, sentence-producing automaton is easy to build, the challenge is to construct one in where a “poetic secondary structure” modulates the outputs.87 Enzensberger puts it, “the production (and understanding) of poetic texts is subject to more than the basic rules of grammar.”88 Enzensberger’s Invitation to a Poemat,89 imagines building a poetry vending machine, an absurdity he uses to think through the existing limits of language-processing. The poemat, Enzenberger feels, is doomed: not simply because “no linguist” could program its language capacities, but also because “poetic competence is a special case of general language capacity.”90 As Enzensberger adduces a certain contradiction in the creation of a truly poetic vending machine, one that would endure the same contradictions as the human user of language: “While the logic of a simple text-generating machine aims for symmetry, grammaticality, redundancy, and monotone, a poemat must attempt a high level of variety, surprise, polysemy, and certain disobedience to rules.” This “secondary structure” of poematicity, he realized, would produce errors in the underlying, rule-based algorithm, such that “the outcome could only be a compromise, one that holds both sides of this dialectic in common.”91 While in 1974, these pronouncements seemed intuitive, by June 2000, the municipal council of Landsberg am Lech actually went about constructing Enzensberger’s poemat for a local poetry festival.92 Computer technology had advanced to the point where this “compromise” of which Enzensberger had written, a compromise between technicity and poeticity, seemed no longer insurmountable.

Gesturing toward the poetics of unreadably massive iterations such as Queneau’s Cent Mille Millards de Poèmes, Enzensberger explains that his Poemat would produce “a not quite endless but indeed quite enormous, perhaps too enormous quantity of poems,” a productivity that would exceed “everything that humanity has brought forth up till now, even if only from a quantitative perspective,” namely, “an order of magnitude of 1036.” The likelihood that the Poemat would emit a repeat permutation of any text, by Enzensberger’s calculations, was “once in 5 x 1029 years, that is, (when considering the astrophysical circumstances): never.”93 Compare this combinatorial fantasy of exhaustive enumeration and poetic productivity with the technological imagery used by Ezra Pound in his 1912 essay, “Psychology and Troubadours.” This machinic imagery is embedded in a material and “germinal universe” of poetry where human mechanisms, “rather like an electric appliance, switches, wires, etc.,” are bathed in a “fluid force” akin to the “glass disc and rotary brushes” or “the wireless telegraph receiver.”94 After the computational era, as N. Katherine Hayles has famously noted, we begin to sustain an altered imaginary where machinic force is replaced by an invisible regime of calculation wherein

Flickering signification extends the productive force of codes beyond the text to include the signifying processes by which the technologies produce texts, as well as the interfaces that enmesh humans into integrated circuits. As the circuits connecting technology, text, and human expand and intensify, the point where quantitative increments shade into qualitative transformation draws closer.95

Poetry critic Brian Reed understands 21st century avant-garde poetics as recapitulating, while expanding, certain aspects of Coleridge’s division between “mechanical” versus “organic” form, in which the rule-based, constraint-driven “mechanical” mode of writing poetry is contrasted to the process-based and spontaneous “organic” form. Reed’s version is articulated as

a new digital era in which, just as all information is reducible to bits, so too all content can be frictionlessly integrated into a horizonless, everlasting present-focused flow. Montage, collage, and juxtaposition were the aesthetic puzzles of the twentieth century. For the twenty-first…the problem is managing the information flood. How do we – can we – inhabit it?96

Enzensberger’s Poemat anticipates and answers Reed’s question by heightening, not lessening, the power of distraction. He envisions the Poemat best placed in a busy airport, where “one has a certain quantum of time” and forced inactivity that leaves one both “attentive” and “passive.” Here, Enzensberger writes, there, “arises a peculiar atmosphere, in which impatience and boredom mix with distractability. These conditions are propitious for play, which needs only short-term attention.”97

For German mathematician and computer scientist Theo Lutz, earliest predecessor to Enzensberger’s more ironic auto-poetic machinations, the purpose of automating poetry focused on the simpler and more pragmatic question, “is it possible?” Canonized as one of the earliest writers of computer-generated poetry in C.T. Funkhauser’s Prehistoric Digital Poetry, Theo Lutz, in 1959, ran an algorithm on a Zuse Z22 computer at the Stuttgart Technische Hochscule that generated stochastic texts – pairs of short, declarative sentences. The program cycled randomly through a restricted repertoire: a set of 16 nouns and 16 predicates. In his write-up of the text-generating experiment, Lutz refers to the code-numbers given to each word as a form of “alphanumeric encryption” [alphanumerische Verschlüsselung] and describes his mechanisms for standardizing agreement between the “truth-values” [Wahrheitswert] of sentences.98 He extrapolates from the foundational ease of working with “unambiguous” [eindeutig] languages, “mathematical language,” to suggest that “with programmable digital computers arises now a synthetic philology, a philology that does not analyze a text but, rather, synthetically generates [synthetisch erzeugt] the text.”99 Lutz places no special emphasis on poetry or poetics in the various articles he published about his algorithm. It seems the experiment has come to be spoken of as digital poetry simply because its first results were published in Augenblick, Max Bense’s journal for contemporary experimental poetics.

Yet, if there is a poetics implied by Lutz’s Stochastische Texte, it is in a possibly unwittingly paradoxical statement Lutz makes concerning the essence of the random generation of language. In the permutating machine, he writes, “unlikely combinations become more likely.”100

This observation regarding the statistical and uncannily unlikely use of language should be read in the context of the “source” text from which Lutz drew his 32-word corpus: The Castle by Franz Kafka. A few examples from the machine’s output read:





In The Castle, as in Lutz’s “stochastic texts,” the probable and the unpredictable overlap. In the narrative of Kafka’s The Castle, a land surveyor arrives at a remote castle only to find that his invitation to survey the baron’s estate is subject to inexplicable delays, technological mediations, and perplexing complications. Kafka’s land surveyor in The Castle adapts to a labyrinthine world in which bureaucratic procedures, sovereign fiat, and meaningless obstacles mutually entangle and thwart each other in nightmarish cycles of deferral. Lutz’s experiment, by randomly regenerating the lexical corpus of The Castle, intriguingly recapitulates not only the atmosphere of the novel (“late,” “stranger,” “old,” “distant,” “dark”) but also reinstates the historical and philosophical query behind Kafka’s textual project: does the “message” hold greater or lesser power in the age of information technology? Kafka’s novel, with its intricate network of telephone calls, form letters, official permits, and hearsay, depicts a ghostly symmetry between the modernized relic of the baronial Castle and the mythic return of the biblical Tower of Babel: its confusion now not simply of multiple languages and codes, but a crisis within informational integrity itself. Even as inscrutable and contradictory messages encircle and confound the land surveyor, this message-proliferation works to diminish any sense that meaningful plans stand behind them or that coherent agents have generated them. Indeed, Kafka’s The Castle is already a novel about the mechanization of message.

Theo Lutz’s randomized Castle (1959), then, represents something more than ancestor for computer- poetry. It is also a technologically generated “answer” to Weaver’s Anti-Babel (1955). Weaver ceaselessly advocated for machine translation as crucial information-age milestone, hoping to leverage a computational theory of languages that “will build part of the way back to that mythical situation of simplicity and power” once possessed by some postulated utopia of prelapsarian monolingualism. Weaver, as spokesperson for language’s computational rewrite, presents the new (or renewed) imperative to reunite all forms of communication into one seamless, measurable, and readable flow of positivities. “Language” was to be replaced by signal processing; all inchoate domains of existence from which “information” could be extracted were to be filtered and standardized for universal computability. As if in ambivalent answer to this Anti-Babel, then, Lutz concocts a new, digital-age mode of lexical confounding that originates in or emanates from the abilities of computation itself. His Stochastic Texts proliferate, complicate, and disseminate Kafka’s The Castle, and these messages—stochastic, yet rule-bound—lead us neither toward a hermeneutics nor toward an intentional speaker.

While the mechanization of Kafka’s novel through the “power and simplicity” of Lutz’s algorithm indeed heralds the dawn of computer-generated literature, if we lose ourselves amidst the festivities surrounding the much-celebrated death of the author we might miss attending to new predicaments for readers: located within a new forest of computer-generated texts that surround us, invite us, permit us, deceive us, and divide us. The Stochastic Texts of Lutz prophesy our now generalized condition in the age of search-engine optimized advertising, ad hoc fake news written by chatbots, and a constant, “benign” background registration of one’s every movement and habit: the “big data” of the infinitesimal and the intimate generated by our position within a technological scaffolding, “the internet of things.” The shadow cast by the Anti-Babel might in fact bear a passing resemblance to the silhouette of Kafka’s Castle. We share in the uncertainties faced by the land surveyor, K., suspended in an indeterminate space of messages that may or may not originate in an author, or a reliable intention; messages that may or may not revoke or permit life and work; accidental messages that the self’s own smallest actions may or may not – stochastically – set into motion.

This background-space of messages in which we find ourselves enmeshed and produced, even as we help to produce it, enables new forms for intelligibility even as it risks disappearance, obsolescence, and loss of intelligibility. In a 2012 article on “The Future of Poetry,” Thomas Wolf meditates on the past and the future of language-generating algorithms and computer-generated literature, concluding that,  “reduced to a simple machine output, poetry would be less like a form of human communication (assuming it ever were so), and more like a birdsong: something we find pleasant as it lilts along in the background.”102 By the parenthetical phrase “assuming it ever were so,” we are meant, it seems, to question whether poetry has ever been a form of communication: a valid query. Yet we might also turn this around: “assuming poetry ever was human,” an open question, in my view. I would suggest that the computational age with its machine-generated texts makes it easier than ever to see that poetry is less an intentional utterance than a nonhuman product of an immanent property of language itself.

From the perspective of most computer scientists, the construction of the Anti-Babel is now nearly completed, as recent announcements concerning Google’s “neural-networked” machine translation successes make clear. Here we find ourselves in a new age of A.I. debates: what immanent properties will emerge from a vast machinic infrastructure built up around the totalizing dyad of “communication and control” as postwar computer scientists and cyberneticists first conceived the computational horizon defining “the animal and the machine”?103 We are asked to prepare ourselves for a poetry demoted to birdsong, “lilting along in the background.” But why, in such posthuman speculations on a poetic background, would humans be left, implicitly, in the foreground? From the perspective of the machines that collect and process the big-data of our daily communications, the question could be reframed: what background are human-generated linguistic acts composing in the process of their real-time accretion? What usable calculations does this background make possible? Are humans, in any meaningful way, in the foreground of these calculations? In our continuing meditations concerning the status the posthuman author, we should not miss an equally important window of opportunity to think through the unpredictable effects of seven-decades-worth of information-age decentering that expressly sought to alter not abstract “humanity” so much as the status of humanity’s languages – shifting the devices of language toward transmission, from intersubjectivity to operational transitivity, from plurality to encryption, and from irreducible multi-lingualism to the “interlingual” computational representations underlying the mathematical abilities of a new generation of machine translation: the next phase of crypto-monolingualism.


  1. Warren Weaver, “Recent contributions to the mathematical theory of communication,” in The Mathematical Theory of Communication (Urbana, University of Illinois Press, 1949). 

  2. Yehoshua Bar-Hillel, “Introduction,” Language and Information (Reading, Mass.: Addison-Wesley, 1964), 1–16. Bar-Hillel held this position at MIT in their Research Laboratory for Electronics. Bar-Hillel was heavily influenced by the ideas of cybernetics and had learned of information theory from Rudolf Carnap while at the University of Chicago’s philosophy department. He saw his appointment as inquiry “a testing ground for the validity of the attempts at formalizing natural languages” as well as effects on international exchange of information See also his early essay Can Translation Be Mechanized?” American Scientist 42, no. 2 (April 1954): 248-260. 

  3. Warren Weaver, “Foreword: The New Tower,” in Machine Translation of Languages, eds. William Locke and A. Donald Booth, v-vii (Cambridge, Mass.: The M.I.T. Press, 1955), vii. 

  4. Peter Galison, “The Ontology of the Enemy: Norbert Wiener and the Cybernetic Vision,” Critical Inquiry 21, no. 1 (Autumn, 1994): 228-266. 

  5. Weaver, “New Tower,” vii. 

  6. Walter Benjamin, “On the Concept of History” [“Theses on the Philosophy of History”] and “Paralipomena to ‘On the Concept of History,’” Walter Benjamin: Selected Writings, Vol. 4 (1938-1940), ed. and trans. Howard Eiland and Michael W. Jennings, 389-411 (Cambridge, Mass.: Harvard University Press, 2003), 389. 

  7. Benjamin, “On the Concept of History,” 389. 

  8. Quoted in John Hutchins, “From First Conception to First Demonstration: The Nascent Years of Machine Translation, 1947–1954. A Chronology,” Machine Translation 12 (1997): 203. 

  9. Norbert Wiener, The Human Use of Human Beings: Cybernetics and Society, 2nd edition, rev. (Boston: Houghton Mifflin Company, 1954), 85. 

  10. Warren Weaver, “Translation,” in Machine Translation of Languages, eds. William Locke and A. Donald Booth. Cambridge, 15-23 (Cambridge, Mass.: The M.I.T. Press, 1955), 18. 

  11. Quoted in Hutchins, “First Conception,” 207. 

  12. A. Donald Booth and William Locke, “Historical Introduction,” in Machine Translation of Languages, eds. William Locke and A. Donald Booth, 1-14 (Cambridge, Mass.: The M.I.T. Press, 1955), 14. They continue: “A machine with a sufficiently extensive storage organ would be able to construct rhymes and rhythms…[if] provided in the memory with suitable routines for processing them. …The act of ‘translation’ would then consist in identifying the ideas contained in the original text and expressing these in terms of stored phrases… a transposition of semantic content… the ‘translation of ideas’ for which the human translator always strives.” 

  13. Claude Shannon, “Prediction and Entropy of Printed English,” Bell Systems Technical Journal XXX, no. 1 (Jan. 1951): 56. 

  14. Rita Raley, “Machine Translation and Global English,” Yale Journal of Criticism 16, no. 2 (Fall 2003): 292-93. 

  15. Raley, “Machine Translation,” 293. 

  16. Michel Foucault, The Order of Things: An Archaeology of the Human Sciences (New York: Vintage Books, 1970). 

  17. For a timeline of computer-generated poetry that begins in 1959 with Theo Lutz’s Stocastische Texte (discussed below), see C.T. Funkhouser, Prehistoric Digital Poetry: An Archaeology of Forms, 1959-1995 (Tuscaloosa: University of Alabama Press, 2007). One particularly infamous machine-poetic imitation-game concerning human poets was conducted by the Forgodot collective in 2008, publishing an enormous anthology of renowned poets (all machine forgeries) online. See the manifesto outlining the philosophy behind the project more generally, as well as a description of the text-generating algorithm: Jim Carpenter, (2004), “Electronic Text Composition Project.” Slought Foundation. [] accessed October 6., 2016. Not all of these poem-generating machines challenge the Turing test in the service of an avant-garde agenda; many NLP/NLG computer science and computational linguistics projects attempt similar feats. See, for example, online poetry forum testing in Slim Abdennadher and Ali el Bolock,“Towards Automatic Poetry Generation Using Constraint Handling Rules,” SAC ‘15 Proceedings of the 30th Annual ACM Symposium on Applied Computing, Salamanca, Spain — April 13 – 17, 2015: 1868-1873. 

  18. Warren Weaver, “Translation [1949],” in Machine Translation of Languages, eds. William Locke and A. Donald Booth, 15-23 (Cambridge, Mass.: The M.I.T. Press, 1955). 18. 

  19. Otto Jespersen, “Interlinguistics,” in International Communication: A Symposium on the Language Problem (London: K. Paul, Trench, Trubner & Co., 1931), 95. 

  20. Jespersen, “Interlinguistics,” 97. 

  21. Jespersen, “Interlinguistics,” 96-97. 

  22. Victor H. Yngve, “Implications of Mechanical Translation Research,” Proceedings of the American Philosophical Society 108, no. 4 (Aug. 27, 1964): 275. 

  23. Booth/Locke, “Historical Introduction,” 2. 

  24. Booth/Locke, “Historical Introduction,” 7. 

  25. The “cryptographic-translation idea” is concerned with the “statistical semantic character of languages” which must “make deep use of language invariants.” See Weaver, “New Tower,” 21-22. 

  26. Anthony Oettinger, “The Design of an Automatic Russian-English Technical Dictionary,” Machine Translation of Languages, eds. William Locke and A. Donald Booth, 47-65 (Cambridge, Mass.: The M.I.T. Press, 1955), 49. He continues: “In telecommunication as in cryptography the arbitrary nature of any given symbolic representation of information has long been recognized.” First attempts in Russia to translate Chinese ideographs used a coding system originally invented for the use of the telegraph. See D. Yu. Panov, Automatic Translation, trans. A. Kirsch (New York: Pergamom Press, 1960), 48. 

  27. Oettinger, “Technical Dictionary,” 49. 

  28. Michel Foucault, “History, Discourse, and Discontinuity [1968],” Salmagundi 20, trans. Anthony M. Nazzaro (Summer/Fall 1972): 229, 230-31. 

  29. Raley, “Machine Translation,” 308. 

  30. Friedrich Kittler, Literature, Media, Information Systems, trans. John Johnston (New York: Routledge, 1997), 160, 158. 

  31. Alan Liu, Local Transcendence: Essays on Postmodern Historicism and the Database (Chicago: University of Chicago Press, 2008), 211. On the topic of XML language with typographic and conceptual categories pre-given in the base code, Liu points to how “the separation of content from material instantiation or formal presentation” is effected by the demands of “transformability, autonomous mobility, and automation” (216). See also Lev Manovich, “The Database as Symbolic Form,” Millennium Film Journal 34 (Fall 1999). 

  32. Viktor Šklovskij, “Parodijnyj roman [1925],” in Russian Formalist Criticis, ed. S. Bann and J.E. Bowlt, 48-72 (Edinburgh: University of Edinburgh Press, 1975). 

  33. Peter Steiner, Russian Formalism: A Metapoetics (Ithaca: Cornell University Press, 1984), 52. 

  34. Guillaume Apollinaire, The Cubist Painters [1913], trans. Peter Read (Berkeley: University of California Press, 2004), 9. 

  35. Apollinaire, The Cubist Painters, 9-10. 

  36. Peter Steiner explands on the wide-ranging uptake of the technological paradigm in Russian Formalism: “The Formalist S. Baluchatyj characterized his method as a ‘technological literary discipline’ …G. Vinokur described stylistics as ‘a kind of ‘linguistic technology’ … B. Ejchenbaum summed up the early phase of Formalism as follows: ‘In recent years, students of literature and critics have paid attention above all to questions of literary “technology”’ (‘Literaturnyj byt” Moj vremennik)” (45n3). 

  37. Roman Jakobson, Selected Writings, Vol. 2 (2 vols) (The Hague: Mouton, 1971), 579. 

  38. Susan Stanford Friedman, “Definitional Excursions: The Meanings of Modern/ Modernity /Modernism,” in Disciplining Modernism, ed. Pamela Caughie, 11-32 (New York: Palgrave Macmillan, 2009). 

  39. Jim Carpenter, “Electronic Text Composition Project.” Slought Foundation. [accessed October 6, 2016], 2. 

  40. Brion Gysin, “Cut-Ups: A Project for Disastrous Success.”  Evergreen Review 32 (April-May, 1964): 60. 

  41. Wiener, Human Use of Human Beings, 79. 

  42. Wiener, Human Use of Human Beings, 15. 

  43. Andrew Pickering, The Cybernetic Brain: Sketches of Another Future (Chicago: University of Chicago Press, 2010), p. 18. 

  44. Pickering, Cybernetic Brain, 26. 

  45. Pickering, Cybernetic Brain, 25. 

  46. John von Neumann, The Computer and the Brain, 3rd ed. (New Haven: Yale University Press. 2012), 82, 83. He argues: “we have now accumulated sufficient evidence to see that whatever language the central nervous system is using, it is characterized by less logical and arithmetical depth than what we are normally used to. [..] Consequently, there exist here different logical structures from the ones we are ordinarily used to in logics and mathematics. They are… characterized by less logical and arithmetical depth than we are used to under otherwise similar circumstance. Thus logics and mathematics in the central nervous system, when viewed as languages, must structurally be essentially different from those languages to which our common experience refers” (82,83). 

  47. Wiener, Human Use of Human Beings,16. Theodor Adorno, writing of the rise of projects like Basic English, closely related to the development of machine translation corpora after the war, fears that developing a language “fit for giving commands,” demonstrates “a communicative ideal that is in actuality an ideal of manipulation; today the word that is designed to be understood becomes, precisely through this process of calculation, a means to degrade those to whom it is addressed to mere objects of manipulation and to harness them for purpose that are not their own.” […] “The universal system of communication, which on the face of it brings human beings together and which allegedly exists for their sake, is forced upon them.” See Theodor Adorno, “Words from Abroad [1959]” Notes to Literature, Vol. 1 (2 vols.), trans. Shierry Weber Nicholsen, ed. Rolf Tiedemann, 185-199 (New York: Columbia University Press, 1991), 191. 

  48. Italo Calvino, “Cybernetics and Ghosts [1967],” The Uses of Literature: Essays, (Harcourt, San Diego, 1982), 6. 

  49. Walter Benjamin, “The Work of Art in the Art of its Mechanical Reproducibility [Second Version, 1936],” Walter Benjamin: Selected Writings, Vol. 4 (1935-1938), ed. and trans. Howard Eiland, Marcus Jennings, and Edmund Jephcott, 101-133 (Cambridge, Mass.: Harvard University Press, 2002), 106, 107. 

  50. Wiener, Human Use of Human Beings, 16. 

  51. Calvino, “Cybernetics and Ghosts,” 24. 

  52. Benjamin, “Work of Art,” 107. 

  53. Gysin, “Cut-Ups,” 59. 

  54. Quoted in Joseph Weizenbaum, Computer Power and Human Reason: From Judgment to Calculation (San Francisco, Calif.: W.H. Freeman and Co., 1976), 201. McCarthy’s comments take place in the context of a BBC broadcast, Second Programme, August 30, 1973 (the ‘Lighthill Debate’ concerning the viability of A.I. 

  55. “the nervous system…is of an essentially statistical character. In other words, what matters are… frequencies of periodic or nearly periodical pulse-trains, etc. Thus the nervous system appears to be using a radically different system of notation … we have here a system of notations in which the meaning is conveyed by the statistical properties of the message… this leads to a lower level of arithmetical precision but to a higher level of logical reliability: a deterioration in arithmetics has been traded for an improvement in logics.” von Neumann, Computer and the Brain, 80. 

  56. Wiener, emphasis added, Human Use of Human Beings, 35-36. 

  57. Wiener, Human Use of Human Beings, 78. 

  58. R.H. Richens and A.D. Booth,“Some Methods of Mechanized Translation [1952],” in Machine Translation of Languages, eds. William Locke and A. Donald Booth (Cambridge, Mass.: The M.I.T. Press, 1955), 35. 

  59. Jacques Derrida, “Des Tours de Babel,” in Difference in Translation, ed. Jospeh F. Graham, 165-207 (Ithaca: Cornell University Press, 1985). 

  60. Michael Cronin, Translation in the Digital Age (London: Routledge, 2013), 21. 

  61. Walter Benjamin, “On Language as Such and on the Language of Man [1916],” Walter Benjamin: Selected Writings, Vol. 1 (1913-1926), eds. Marcus Bullock and Michael W. Jennings, 62-74 (Cambridge, Mass.: Harvard University Press, 1996), 61. 

  62. Jakobson, Selected Writings, 569. 

  63. Jakobson, Selected Writings, 579. 

  64. Jakobson, Selected Writings, 575-76. 

  65. Jakobson, Selected Writings, 574. 

  66. Jakobson, Selected Writings, 578. 

  67. Jakobson, Selected Writings, 574. 

  68. Solomon Marcus, “Linguistics as a Pilot Science [1963],” in Current Trends in Linguistics, Vol. 12 (14 Vols.), ed. Thomas Sebeok, 2871-2888 (The Hague: Mouton, 1974), 2872. 

  69. Marcus, “Linguistics as a Pilot Science,” 2880. 

  70. An important dimension of this is economic: linguistic historian John Earl Joseph, writing on post-World War II “algebraic’ structuralism,” outlines how “mathematical aspects of structuralism, in the use of tables, formulas, and other mathematical schemata, statistics, calculations” bent linguistics toward systematicity, as “military and commercial interest in furthering the wartime progress on computers and machine translation” economically incentivized American linguists to pursue computationally-based models. See John Earl Joseph, From Whitney to Chomsky: Essays in the History of American Linguistics (Amsterdam: John Benjamins Publishing Company, 2002), 60. 

  71. Emily Apter, The Translation Zone: A New Comparative Literature (Princeton: Princeton University Press, 2006), 227. Pioneers in the formalization of human language include linguists Otto Jespersen and Leonard Bloomfield, among others: “As to denotation, whatever can be said in one language can doubtless be said in any other: the difference will concern only the structure of the forms, and their connotation.” In Leonard Bloomfield, Language (New York: H. Holt and Company, 1933), 278. “An ideal language would always express the same thing by the same, …any irregularity and ambiguity would be banished.” See Otto Jespersen, Progress in Language: With Special Reference to English (London: Swan Sonnenschein & Co., 1894), 365. 

  72. Zellig Harris, “Review of Selected Writings of Edward Sapir in Language, Culture, and Personality,” Language 27 (1951): 301. 

  73. Weaver, “Translation,” 21-22. 

  74. Weaver, “Translation,” 23. 

  75. Kittler, Literature, Media, Information Systems, 148. 

  76. Slim Abdennadher and Ali el Bolock,“Towards Automatic Poetry Generation Using Constraint Handling Rules,” in Proceedings of the 30th Annual ACM Symposium on Applied Computing (SAC’15)(2015): 1870. 

  77. Benjamin, “Work of Art,” 118-119. 

  78. Benjamin, “Work of Art,” 118-119. 

  79. Andrey Bely, “The Principles of Aesthetic Form [1909],” Selected Essays of Andrey Bely, ed. and trans. Steven Cassedy (Berkeley: University of California Press, 1985), 217. 

  80. Gertrude Stein, Writings and Lectures, 1909-1945 (London: Penguin, 1967), 126. 

  81. Ferdinand de Saussure, Cours de linguistique générale: Tome 2, Appendice: Notes sur la linguistique générale, ed. Rudolf Engler (Wiesbaden: Harrowitz, 1974), 22. 

  82. Calvino, “Cybernetic Ghosts,” 10. 

  83. Joseph Weizenbaum, Computer Power and Human Reason: From Judgment to Calculation. (San Francisco: W.H. Freeman and Co., 1976), 193. 

  84. Weizenbaum, Computer Power, 193. 

  85. Paul De Man, “Conclusions: On Walter Benjamin’s ‘Task of the Translator’ [1983],” The Resistance to Theory (Minneapolis: University of Minnesota Press, 1986) 84. 

  86. Hans Magnus Enzensberger, (1974) Invitation to a Poemat [Einladung zu einem PoesieAutomaten] (Frankfurt: Suhrkamp, 2000), 19. All translations from the German are my own. 

  87. Enzensberger, Invitation [Einladung], 29-30. 

  88. Enzensberger, Invitation [Einladung], 18. 

  89. I’ve translated “Poesie-Automaten,” from the original German title, as “Poemat” to preserve the etymological sense of “self-moving” that the concrete association of “Automaten” with “vending machine” does not convey. 

  90. Enzensberger, Invitation [Einladung], 18. 

  91. Enzensberger, Invitation [Einladung], 31-32. 

  92. Also used in 2006 to commentate on World Cup football matches, the poemat can now be found in the Literaturmuseum der Moderne (in the Deutschen Literaturarchiv Marbach). 

  93. Enzensberger, Invitation [Einladung], 22. 

  94. Ezra Pound, “Psychology and Troubadours, Quest 4, no. 1 (May 1912): 44-45. 

  95. N. Katherine Hayles, How We Became Posthuman: Virtual Bodies in Cybernetics, Literature, and Informatics (Chicago: University of Chicago Press, 1999), 46-47. 

  96. Brian M. Reed, Nobody’s Business: Twenty-First Century Avant-Garde Poetics (Ithaca: Cornell University Press, 2013), 87. 

  97. Enzensberger, Invitation [Einladung], 34-35. 

  98. Theo Lutz, “Über ein Programm zur Erzeugung stochastisch-logistischer Texte,” Grundlagenstudien aus Kybernetik und Geisteswissenschaft 1, no.1 (January 1960): 12, 13. 

  99. Lutz, “Über ein Programm,” 14, 11. 

  100. Lutz, “Über ein Programm,” 16. 


  102. Thomas Wolf, “The Future of Poetry,” Opticon1826 Vol 3 (2012): 3. 

  103. Norbert Wiener, Cybernetics, or Control and Communication in the Animal and the Machine (Cambridge, Mass.: The M.I.T. Press, 1948). 

This essay was written during a postdoctoral fellowship at the Penn Humanities Forum. I am grateful to all my colleagues in the Forum’s weekly workshop, who read and commented on earlier drafts. I am especially indebted to the guidance, insights, and generous conversation of James English, Michael Gordin, Judith Kaplan, Laura Kunreuther, Rahul Mukherjee, Sheila Murnaghan, and Christine Poggi.
Article: Author does not grant a Creative Commons License as part of this Agreement.