Archive for August 21st, 2008

1964: Baudot, La machine à écrire

August 21st, 2008 — 06:25 pm

1964: Jean Baudot, a pioneering engineer-linguist, creates the first French machine-generated published poetry.

Published by the Editions du Jour in Montreal,”La machine à écrire mise en marche et programmée par Jean A. Baudot” (“A Writing Machine created and programmed by Jean A. Baudot”) is still circa 2008 available (mildewed and seemingly unread since 1976) in the Concordia library. A rough translation below (by myself) of Jean Baudot’s introduction reveals his language and concerns as strikingly contemporary and lucidly clear. Either time has stood still or it seems that new media evoked unresolved concerns early in its evolution.

In this preface to his pioneering work on human-machine creativity Jean Baudot, who was an engineer by training and became a linguist writing on formal grammars, writes about the ubiquity of computers, their capacity to emulate human tasks, and his goals.

Humans have always been attracted to automation. From the beginning of time, humans have invented devices to imitate and surpass human capacities. Most often these machines have reassured humans of some control over the material world.

Certainly a sense of domination is elicited when contemplating a machine performing a task previously only possible through labor. We find ourselves stronger and above all conscious of our privileged nature.

Technological development of recent decades has taught us to be astonished by the power of machines. We know that machines are work tools. Its with them that we progress.

In this domain , computers – loosely called electronic brains – play a major role. These utilities have invaded industry. Without them a big part of our scientific, industrial and commercial activities would be instantly paralyzed. In effect, computers, vast manipulators of data and info, can be utilized for executing very varied tasks. They are model students. It is sufficient to show them correctly, only one time, how to execute a task for them to accomplish it at often prodigious speeds. One such machine can learn a multitude of different tasks, and always remembers the particularities of each.

The phrases which appear in this volume ["La machine à écrire mise en marche et programmée par Jean A. Baudot" ] were composed by a computer. The texts are less a literary performance, but more the result of an experience which merits some interest. Composition is considered, without any doubt, as fundamentally human activity, it is therefore troubling to observe a machine functioning without any external intervention writing evocative phrases in a credible style. How can it be possible? It’s extremely simple. It is sufficient to teach the machine some grammatical rules, a foundation vocabulary and let it work. We assist then the works of a genuine robot which writes without comprehending what it says because it doesn’t know the sense of words. [...]

Our goal was to observe how a machine behaves after it has been taught a little grammar and has at its disposal a constrained lexicon (630 words approximately). In order to avoid introducing, consciously or unconsciously, bias taken in the choice of words placed at the disposition of the computer, we decided to extract a manual of French of the simplest level possible.

To that end we chose the manual of the 4th year actually used in our schools and entitled “My French Book” (Brothers of the Sacred Heart). The 630 corpus represents about half of the words utilized in the manual. All the words utilized are therefore simple and at the level of a 10 year olds vocabulary.

During the research, the machine having been appropriately programmed was left running overnight. Imagine our surprise the next morning to discover it had printed thousands of phrases and it seemed as if it could continue without stopping. This volume represents a ample of those phrase composed by automated processes. The phrases are reproduced as they appeared, even if sometimes the temptation was strong to modify them slightly.

I leave it to the reader, literrati or amateur of new styles, to their own conclusions.

Jean A. Baudot
Montréal, juin 1964
p.s. the reader, interested in the technique related to this automated process, will find some explanations in the appendix.

But what of the poetry created by Baudot’s machine? Baudot warns us to consider it as examples of a process not a literary exercise. And that is an appropriate warning because the text is only occasionally luminous and as fragmented as a drunk HD trying to smoothly waltz. It’s also a bit like a randomized scrabble board played by semi-literate spiders: the sentences are stiff formal aphorisms that never congeal into sustained impact. It possesses astonishingly readable basic grammar but is lacking in the subtle contours of emotional play and emotional taste of life. These are machine words. Fragments that suggest a state space of potentialities that marches and meanders toward automated plot-generators and Kurzweil’s Cybernetic Poet.

Shown to a Quebec visual-artist using the pretense that they were poems by a human, the language of the machine-generated poems immediately evoked Baudelaire and Alain Robbe-Grillet. The reader responded: “I don’t understand the juxtaposition of words… Other poetry has a flow that I can feel and understand. This I can’t.”

These alienated responses resonate with experiments done by linguists in natural language: humans cannot learn artificial languages without effort. Generative grammar suggests a neurological foundation etched into synaptic circuitry that predisposes us to syntactical conjunctions and organic morphemes. Extrapolating, perhaps there is a neurological parser for art, a dendrite module for meaning, a cluster coiled into a knot experienced as soul. Similar speculations have motivated the search for the neuronal correlate of consciousness by Christof Koch and Francis Crick.

In one of the appendices to La machine à écrire the quebec poet-troubadour Felix Leclerc points out a crucial ongoing often-repeated unresolved challenge to computational creativity:

“Ask it [the computer] to be numbers, that’s reasoning, it will be it, but to be heart, I don’t believe it.” ~ “Demandez-lui d’être chiffre, c’est-à-dire raison, elle le sera, mais d’être coeur, je ne le crois pas.” [p.75]

Subtle contextual connectivity, the sinew of narrative, the sinuous twisting truth of lived emotional reality, the ache and ebb of our tidal hormonal interiorities: even contemporary computer-created art lacks this sensitivity. The reason remains the same as what Baudot clearly states: computers are not aware of meaning, the computer “doesn’t know the sense of words. “. (Human meaning at least.) Lack of flow remains a central flaw. Here are a few sentences in both english and french from page 45 of “La machine à écrire”:

La vacance et un mari oublieront des fillettes. ~ The holiday and husband will forget girls.

Une peur cultive un serpent. ~ Fear cultivates a snake.

The meanings that emerge from these phrases are imported by the reader: snake and fear accidentally bumping against each other in the archetypal basement provoke a tiny spark. The result is unintended and so resists integration into a sustained sense of the work as art. One could argue that decades of artists following in the footsteps of John Cage have elaborated complex strategies for including process and contingency into art-making. But behind those products, the human artist acts as conceptual filter, explicating and enriching approaches to accidents by placing them through discourse within historical context.

Baudot’s work is contextually a process-artwork that highlights the dilemma of meaning. When computers become conscious of meaning (which involves all the attendant emergent psychic tendrils of purpose intuition and need) then perhaps machine-created art will become meaningful in a way capable of sustained emotive interest rather than intellectual curiosity. As it is Baudot’s work is a crucial preliminary step which anticipates the core of generative poetics practice as it has continued for the intervening 4 decades since 1964: grammars and recombinant structure.

One tendency of contemporary computer-created art-work is to circumvent or sublimate this deficiency of meaning flow (and the lack of an ineffable taste of an auteur’s predelictions in machine-created output) by investigating low-fi DIY aesthetics and conceptual interventions in the hope of distracting viewers from the essential reality that emotional depth remains computationally intractable.

2 comments » | Uncategorized