Craft and Automation

Abstract

How far has the Oulipo gone in the directions of conceptual writing and automation? Conceptualism is built into the very idea of potential literature, but the group’s interest in automation has gradually waned. Computer-assisted literary composition was delegated to an offshoot workshop (the Alamo), which has not thrived. If the Oulipo's activity is essentially craftlike, as Jacques Roubaud has written, this is not surprising. Nevertheless, some of the group’s members have used their craft to imagine or simulate automation. Georges Perec and Marcel Benabou’s P.A.L.F. (“Production automatique de littérature française”) project, and Perec’s radio play Die Maschine illustrate this strategy.

Keywords:

• Oulipo
• conceptualism
• automation
• Marcel Bénabou
• Georges Perec

In “The New Writing” (1998) the Argentinean novelist César Aira, disputably identified by Veronica Esposito as a true inheritor of the Oulipo, claims: “the great artists of the twentieth century are not those who produced a body of work, but those who invented procedures so that works could make themselves, or not” (Elkin and Esposito 38–41, “La Nueva escritura”). Procedures, Aira implies, open up two avenues: automatically produced art and conceptual art. The procedure can be implemented automatically, or a formulation of the procedure can stand in for the artwork. How far has the Oulipo gone in these two directions? The short answer is: further in the conceptual direction. Which is hardly surprising, since the conceptual is built into the foundational notion of potential literature, enshrined in the group's name.

In their privileging of the constraint and the procedure, and in their fondness for describing imaginary, unmade artworks, the writers of the Oulipo have enduring affinities with conceptual art. But with the exception of Marcel Bénabou’s exploitation of langage cuit (sayings and proverbs), they have shown only a moderate interest in what for Craig Dworkin is the central feature of contemporary conceptual writing: “the use of found language in ways that go beyond modernist quotation or postmodern citation” (xliv). And although the Oulipo has an indispensable website, it has not reacted to the appearance of the internet as to a given that is, according to Kenneth Goldsmith's analogy, revolutionizing literary writing just as photography changed the game for painting.¹

What of the other direction signalled by Aira’s manifesto-essay: automation? Some progress towards the automatic production of literary texts was made in the first two decades of the Oulipo’s activity. Queneau’s Cent mille milliards de poèmes (A Hundred Thousand Billion Poems, 1961) and his branching “Un conte à votre façon” (“Story of Your Own,” 1973) resemble plans for combinatorial machines.² Although originally published on paper and requiring manual manipulation, they were quickly automated. When the Alamo (Atelier de Littérature Assistée par la Mathématique et les Ordinateurs, or Workshop for Literature Assisted by Mathematics and Computers) was founded in 1981 by Paul Braffort and Jacques Roubaud as a “computing arm” (“prolongement informatique”) of the Oulipo, automation was naturally a part of its brief (Alamo). But the design of an automatic system that can produce interesting literary texts is, of course, a very complex task, and without sustained industrial or institutional backing, neither the Oulipo nor the Alamo was ever really in the race. As Jacques Roubaud said in an interview with Camille Bloomfield: “computing is very expensive, and Oulipians are not big spenders” (Bloomfield 248).

The Oulipo’s recent inactivity in computational creativity has been a disappointment to some, but is not entirely surprising. Apart from the external reasons already mentioned, there has been a certain lack of engagement, notably on the part of the Alamo’s co-founder, Jacques Roubaud. According to Paul Braffort, Roubaud was never really interested in the Alamo (Bloomfield 250). A paragraph from “La Mathématique dans la méthode de Raymond Queneau” foreshadows that disinterest:

Le travail oulipien est artisanal.

Le commentaire de Queneau semble ici un masque : … “ceci n’est pas essentiel, nous regrettons ne pas disposer de machines.” Je l’entendrais un peu différemment ; et les machines ne changerait rien à l’affaire. Il me semble […] qu’il s’agit d’un trait, au contraire, essentiel. La revendication de l’artisanat renvoie à l’affirmation de l’amateurisme ; est un archaïsme volontaire (et peut-être, ici aussi, une anticipation). (Oulipo, Atlas 53)

This is not necessarily a disavowal of the Oulipo’s original brief, which privileged mathematics over other potential complementary disciplines (such as linguistics). As the eminent computer scientist and mathematician Donald Knuth has written: “there is no such thing as ‘mathematical thinking’ as a single isolated concept; mathematicians use a variety of modes of thought, not just one” (180). And not all mathematicians prefer to use the modes most suited to computer science and algorithmics.

Because of the graphic complexities of mathematical notation, many mathematicians, even today, spend less time at computer interfaces than their colleagues in the humanities. Andrew Wiles, who proved Fermat’s last theorem in 1995, was reported to have used a computer only to write up his results.³ In 2002, the Fields medallist Timothy Gowers wrote that most of his colleagues did not use computers in their work in a fundamental way.⁴ This is changing, as Gowers believed it would. Computers are used systematically to verify proofs in the field of experimental mathematics, which has had its own journal since 1992.⁵

Automation, then, is advancing in mathematics as it is in natural language processing. But individuals and groups may have reasons for not contributing to its advancement. To dismiss such reasons automatically as “innovation resistance” is to assume that automation has a manifest destiny to spread into all areas of human activity. As Quassim Cassam has pointed out: “Many innovations have turned out to be not fit for purpose and those who resisted their adoption have been proved right in the end” (1). If tasks are neither boring nor onerous, there might be good reasons not to automate them, even if we can. Crafts, after all, have made something of a comeback, and the Slow movements look set to last for as long as the destructive logic of productivism keeps pushing us in the opposite direction.

If the Oulipo’s activity is essentially craftlike, as Roubaud argued, it is not surprising that the group has not become a leader in automation, for craft and automation are fundamentally at odds in certain key respects. A simulation of craft production can be automated, but craft itself, by definition, cannot. Inversely, automatic production must, by definition, run without human intervention, but a simulation of it can be crafted. A famous example is Wolfgang von Kempelen’s chess-playing “Turk:” a false automaton, operated by a hidden chess master, which defeated numerous challengers in Europe and America from 1770 to the early nineteenth century. In literature as in live entertainment, machines may have imaginary uses, as science fiction abundantly shows. Some members of the Oulipo have used their craft to simulate automation, not merely imagining machines, but presenting the hand-made as the machine-made, building literary “mechanical Turks.” Marcel Bénabou and Georges Perec’s “Production automatique de littérature française” or P.A.L.F. project, which was their letter of introduction to the Oulipo, is one such invention.

“Automatic production” can mean a variety of things, depending on the kind of mechanisms understood to be driving the process: physical, psychological or linguistic. Bénabou and Perec appeal to the third kind of mechanism, explicitly rejecting psychological automatism as exemplified by surrealism: “La notion d’automatisme n’a aucun rapport ici avec l’écriture dite ‘automatique’ de Breton” (25–26). Breton explicitly acknowledged his debt to Freud’s free-association technique, and defined surrealism in the first manifesto as “psychic automatism in its pure state” (26). Bénabou and Perec, writing in the time of structuralism and generative grammar, appeal to a new paradigm. But can the two kinds of mechanisms (psychological and linguistic) be so sharply dissociated? Jacques Lacan, whose name figures in the “draft table of contents” for the P.A.L.F. (1967) as the author of a projected “letter / preface,” would not have thought so (98).

The P.A.L.F. project was manual and time-consuming, as Bénabou reveals in his introduction to the dossier. It was never designed to be mechanizable. The procedure consists essentially of replacing lexical words in a given sentence with definitions chosen from Littré’s dictionary, as in the following example:

La marquise sortit à cinq heures.

marquise: toit avancé soutenu par des piliers

sortir: être mis en vedette

cinq heures: l’heure du thé

Donc, “La marquise sortit à cinq heures” = le toit avancé soutenu par des piliers fut mis en vedette à l'heure du thé

toit: couche supérieure (d’un filon)

avancé: qui touche à sa terme

soutenu: sans familiarité

pilier: fourche patibulaire

être mis en vedette: être en sentinelle

à l’heure: exact

thé: (arbrisseau) qui croît à la Chine

Donc, “La marquise sortit à cinq heures” = La couche supérieure qui touche à son terme sans familiarité avec les fourches patibulaires est la sentinelle exacte qui croit à la Chine (27)

Human choice intervenes at two levels: choosing among dictionary definitions, and making various adjustments to maintain coherence. The definitions chosen are obviously not the most obvious. Bénabou and Perec declare that they are seeking the greatest possible “automatism” and the greatest possible “diffraction.” They maximize “diffraction” by choosing the definition in the semantic field most remote from that of the word in its initial context.

As Bénabou and Perec note, the procedure quickly overloads the structure of the initial phrase, which is allowed to collapse: “La structure syntaxique de l’énoncé initial ne constitue pas une contrainte absolue; seule s’impose la nécessité de suivre rigoureusement l’ordre de succession des mots,” by which they mean the lexical words (28). Thus “soutenu par des piliers” becomes “sans familiarité avec les fourches patibulaires,” avec replacing par to ensure a minimum of semantic coherence. The simple past tense of “fut mis en vedette” is shifted to the present: “est la sentinelle.” Otherwise the verb croire would have had to be in the imperfect tense (“qui croyait à la Chine”), which would have eliminated the homonymy croît (grows) / croit (believes), on which the final substitution hinges. The word arbrisseau in Littré’s definition has been simply left out.

The examples of their procedure that Bénabou and Perec give have been carefully handcrafted. The basic point of the exercise is to demonstrate that any two lexical items in French are linked by a chain of synonymic or homonymic substitutions passing via subsidiary meanings, and that by the same procedure any two strings of comparable length can be made to converge on a third. Thus Bénabou and Perec show that the punning first and last sentences of Raymond Roussel’s story “Parmi les Noirs”—“Les Lettres du blanc sur les bandes du vieux billard” and “Les Lettres du blanc sur les bandes du vieux pillard”—are “rigorously equivalent” since they can both be transformed into the sentence: “Les bandes de la lettre sur les pillards du vieux blanc” (40–41).

Bénabou and Perec worked intermittently on the P.A.L.F. from 1966 to 1973. As Bénabou writes in his introduction, “finish the P.A.L.F.” became a catch phrase in their conversations and letters, but the projected content kept changing, and eventually frustration at the delays gave way, on Perec’s part at least, to “scepticism and lassitude” (7). Having spent many hours mining the Littré and produced a handful of witty exercises demonstrating the procedure and variants thereof, the point seems to have been made, and the ambitious plans for a “semantic map” of the French language and a novel based on the transformation of “Prolétaires de tous les pays, unissez-vous” into “Le presbytère n’a rien perdu du son charme ni le jardin de son éclat” were abandoned (98). Somewhat ironically, the “Automatic Production of French Literature” faltered and stalled (22).

The P.A.L.F. system produces literature not from raw materials but from fragments of finished works. This is part of the point: literature is made by transforming what has already been written; like language, it “goes round in circles and functions in a closed loop” (9). In the context of the P.A.L.F., the operators of the circulation were dictionary-manipulating human beings, but while that project was still a going concern, Perec embarked on another, which simulated an autonomous machine transforming a canonical work. This was the radio play Die Maschine (The Machine), translated by Eugen Helmlé in close and creative collaboration with the author for the German broadcaster Saarlandische Rundfunk.

In Die Maschine, Goethe’s famous “Wanderers Nachtlied II” (“Rambler’s Lullaby II”) is analyzed, manipulated and finally related to other poetic texts by a machine consisting of three processors responding to “system control.” The machine works through a series of protocols: statistical, linguistic, semantic, critical and poetic. Although comic and desacralizing effects predominate in the sections governed by the first three protocols, there are uncanny and uncomfortable moments. These result from apparent dysfunctions. “Aleatory recreation” misfires and recreates other texts:

üb üb

üb immer treu und redlichkeit

stop

[…]

überalle

überalle

über alles

in der welt

TON (17)

“Üb immer Treu und Redlichkeit” (“Always practice fidelity and honesty”) is the first line of a song by Ludwig Christoph Heinrich Hölty, often cited as an encapsulation of the “Prussian virtues.” From there, the processors drift to the second line of the Deutschlandlied, Germany’s national anthem since 1922, although only the third stanza has been sung on official occasions since the Second World War, the others, and especially the first two lines (“Deutschland, Deutschland über alles, / über alles in die Welt”) bearing the taint of association with the Nazi regime. Ulrich Schönherr's translation of Die Maschine adds the word germany to conserve the salience of the allusion for English-speaking readers:

overall

above all

germany above everything

in the world

SOUND (45)

A little later, system control intervenes just in time to stop processor 2 completing a reflex action from a former time:

h

ha

hai

hai

heil hi

stop (“The Machine” 49)

While they betray traces of all-too-human history, the mistakes in Die Maschine function in a way that Perec intended to be specifically machine-like. He was fond of Paul Klee’s saying, “Genius is the error in the system,” which he quoted when explaining the necessity of the clinamen: “Chaque fois qu’on veut appliquer rigidement un système, il y a quelque chose qui coince. Pour qu’on puisse fonctionner dedans avec liberté, il faut introduire volontairement une petite erreur” (Entretiens I 240–421). But an error deliberately introduced is an error in relation only to the system, not the writer’s intentions; it is a designed irregularity. The truly accomplished writer, Perec implies, is the one who knows how to dispose such irregularities with skill. It was in precisely these terms that he began to think about his radio play. In a letter to Helmlé, he wrote:

To begin with I wanted to explore the relationship between system and error (since genius is the error in the system). First I thought: This is where poetry lies. Then it occurred to me that the genius of a machine is the precise opposite—to be a system based on error. (cited in Bellos 380)

This is a neat conceptual chiasmus (from the error in the system to the system in the error), which reinforces the distinction between human and computer (“so there would be no intersection at all between the two kinds of genius,” writes Perec (cited in Bellos 380)). It is hard to discern, in the finished radio play, a unitary “system based on error,” but there are certainly significant patterns in the products of supposedly random or undirected processes.

The P.A.L.F. project was presented, albeit light-heartedly, as research. Bénabou and Perec claimed that the procedure was “entirely and really” automatic (22). Die Maschine, by contrast, presents itself from the outset as a simulation: “This radio play seeks to simulate the functioning of a computer programmed to analyse and decompose Johann Wolfgang von Goethe’s ‘Rambler’s Lullaby II’” (“The Machine” 33). It is a work of science-fiction. Simulating an erratic automaton gave Perec great freedom in choosing how the analysis and decomposition would unfold. The work was commissioned, promptly completed, and went on to enjoy remarkable success on the airwaves (Bellos 379–385).

Christelle Reggiani has argued persuasively that the Oulipo’s mathematization of literary language has largely been an enabling fiction.⁶ Similarly, the group’s most fruitful engagements with automation have, like Die Maschine, been imaginary. Nevertheless, they have borne real fruit. The group’s practice, particularly in works as opposed to exercises, does not line up with César Aira’s declaration in “New Writing” that the great artists of the twentieth century are not makers but inventors of procedures by which works can make themselves or not (“La Nueva escritura”). That claim, of course, can be disputed if the realized artwork still counts for something. Aira himself, incidentally, has never employed automatic or semi-automatic procedures (as he has admitted in an interview), except in the surrealist sense.⁷ He drafts his novels by hand with a fountain pen. This inconsistency should not come as too much of a surprise. Creative writers, after all, specialize in imagining, and imagination, as Berys Gaut reminds us, “does not involve a commitment to performing an achievable action” (160). Which is not to deny that it can foreshadow actions that will be effectively achieved.

Additional information

Notes on contributors

Chris Andrews

Chris Andrews teaches at Western Sydney University. He is the author of Poetry and Cosmogony: Science in the Writing of Queneau and Ponge (Rodopi, 1999), and Roberto Bolaño’s Fiction (Columbia UP, 2014). He has also translated books of prose fiction, including César Aira’s Shantytown (New Directions, 2013) and Kaouther Adimi’s Our Riches (New Directions, 2020).

Notes

1 See Kenneth Goldsmith, “Why Conceptual Writing? Why Now?” Against Expression: An Anthology of Conceptual Writing, edited by Craig Dworkin and Kenneth Goldsmith, Northwestern UP, 2011, pp. xvii–xx [pp. xvii–xxii].

2 See Raymond Queneau, Œuvres complètes I, edited by Claude Debon, Paris, Gallimard, 1989.

3 See Simon Colton, “Computational Discovery in Pure Mathematics,” Computational Discovery of Scientific Knowledge, edited by Sašo Džeroski and Ljupčo Todorovski, Berlin/Heidelberg, Springer Verlag, 2007, p. 196 [pp. 175–201].

4 See Timothy Gowers, Mathematics: A Very Short Introduction, Oxford UP, 2002, p. 134.

5 See Ian Hacking, Why is there a Philosophy of Mathematics at All? Cambridge UP, 2014, pp. 63–66.

6 See Christelle Reggiani, Poétiques oulipiennes: La contrainte, le style, l’histoire, Genève, Droz, 2014.

7 See César Aira, “Leyendo novelas no se aprende nada,” Interview with Javier Rodríguez Marcos, El País (Spain), 24 June 2016, https://elpais.com/cultura/2016/06/23/babelia/1466689420_025152.html. Accessed 24 Sep. 2020.