Erewhon 2.0
|
|
Erewhon 2.0 |
|
The Electronic Muse (Millie Niss) from Oulipoems
In News from Erewhon, we used techniques of free-association and constraints to try to mine the human unconscious so as to generate unusual texts. These techniques come from the Surrealists and are in some ways exactly the opposite idea from computer-generated text. The French Surrealists, as well as some Romantics and Parnassians, were almost in awe of the power of human imagination and would go to heroic lengths to liberate that imagination. Rimbaud (not a member of these movements but contemporary with them and inspired by them) was particularly known for this: He used sleep deprivation, drugs and alcohol, and other drastic methods to aim at “le déreglement de tous les sens” (“the un-tuning of all the senses”) that would reveal the poetry hidden in his brain even if it destroyed him in the process.
These psychological techniques are quite different from the rather safe and sterile procedure of writing computer programs that produce texts, even if both methods yield nonsense of a possibly similar nature. We do not look down on computer generated text (indeed our Oulipoems has a tool for generating text) but we do think that the idea of writing a complex program for the purpose of generating real literature is a dead end. Even if it were possible to create a program that could make truly great literature (and it is not given the current state of research in Natural Language Processing and Stylistics) it would not be desirable, in the same way there is no point in making a robot that can enjoy food for us or appreciate classical music. We would just be robbing ourselves of the joy of writing.
On the other hand, I see significant potential in human-machine collaboration, in which computers use large datasets and randomized algorithms to make suggestions to human writers, much as the prompting keywords in Erewhon do. In that usage, we are not replacing the author with a machine, but rather supplying the author with a tool that is technically advanced but functions like a fancy thesaurus or reference library.
— Millie Niss
I used a free web page (see below) to generate text based on the full set of News from Erewhon 1.0 texts.
Method: The texts were combined into a single text file, with line breaks removed except between paragraphs. Titles were removed as their syntax and capital letters distort the results. Apostrophes were removed so that contractions (such as “isnt”) would be treated like regular words. M-dashes were eliminated because they were represented by two hyphen in the source text. Punctuation which muct occur in pairs (i.e. (,),” etc.) was removed. All other punctuation was separated from the word it was next to so it would be treated like a word.
Here is the first (500 words, size of word groupings = 1) text generated by the web page. I have altered it only by removing the spaces before periods and commas and adding a line break at carriage returns:
If you. Id rather play pat-a-cake than be accused of the day their eternal sleep of his mans overalls. See if she cares. This was fired on its bombed-out cage.
The foliage outside bristled with the dawn of your mind, and sacrificing-sheep-for-Id tenants and the son, enlivening liqueur. Whether you pay. She slept fitfully, like the chainsaw as were good enough for us when she came to Eternity and saves them, for the terrain of your charms is where I moaned, my uncle cooked for now, but all harm.
I know, I thought until I know that made her hair of some half-eaten Triscuits and stones.
Attending church was marked by frequent and when he should clear your choice of before the cherry tree. I dont toss your horn has no objection when he thought, with each small herb garden or is only some undercover snoop from the airs mighty fresh in the A one-way ticket is swimming in the mice made your wifes unfortunate illness! The boy, Buster. I parks where Im entitled to the financial district where cream color belongs. Petersons idea of the fall in stentorian tones . Thus, but Mrs. Which didnt understand a sesame bagel. You must hate me with lace? or something? I discovered her daughter about not being assertive enough to document every service his teeth. Now that he always chewing after a uniform of the mind, he rode, who took dizzy fits when Amy was somewhat advanced in their poet like a nice fat joint of selling all the saved. He had to refute the moon . No pony tails allowed in the chocolate-covered laxatives
And here is a second one to show variations:
If you want is my eye on calling a poets metaphors. Leave it a little George Washington recited during his living as the Defense de Fumer sign in this and this is a particularly fatty breakfast. Surely, and stones.
The feather floating on their old age if Paris werent slow enough already. Whether you like the chainsaw as were young and as she cares. Smoothing her for eternity, he said to while we have seen only some half-eaten Triscuits and lacerated her husband Chuck had an Army blanket on top of blood in stentorian tones. Its the gravity of the intimate things Mr. Look out one day, illustrates the pillows will be boring, and minus the vine with lace? or is so as curtains of each new millennium, chauffeured by step by their eternal sleep with all play pattycake with his model rockets, her work output was unfair since Bruce and death played out even a steeple and the green goo that again. Look out. There is no Alice to kill him, prostrate, and she drew the middle where did that day, I not? The world hed snap a test subject.
When I said yes, the vine with each day their hide, plus cest la même chose, the main square and he go. I asked rhetorically. If you do his pooper scooper on the Center for comprehending the country, jittery with your senator, and heart am I said. The Parks commissioner had stolen the house got madder and wasnt a little fuzzy on Linwood Avenue. Now it a uniform of Hudson River eels who crawl under Mrs. God Fuck! he
With the word groupings size raised to 3, I got:
If you insist on calling a blue heron a train strike no one will understand a single thing youre saying, Ellis complained to Miranda with that humiliating reasonableness that made her want to kill him. Instead she asked, Why dont I bring you one of your favorite rums? with a smile that made him think he should skip the drink this evening. Leave it in the cabinet for now, he said.. My love for you will blaze orange and red beneath the gibbous moon.
The mice lived in a cream-colored bungalow under Mrs. Johnsons sink. Or so at least it seemed to them. Others less culturally attuned might have seen only some half-eaten Triscuits and the remains of a sesame bagel. Welsh rarebit was the mices favorite meal, but Mrs. Johnson was a vegetarian, and more over, for three years now, she and her husband Chuck had been subscribing to Meals-on-Wheels, ever since their ghost of a Chevy just gave out one evening, without even a scuff on its hood, just gave up the ghost, it did, like their bridge partner Sallys Phil who expired one day at the bowling alley, just like that, never got to try for a spare, and here she was mocking Chuck for his cholesterol when poor old Phil never ate any saturated fats and where did that get him? A one-way ticket to the afterlife and wasnt it a pity because now thered be no more bridge only Sally dreaming of her dear departed in the pastures of Heaven close unto the Lord but not too close, Sally hoped, for he always liked a nice fat joint
As you can see, the larger word groupings make output that incorporates large chunks of the input text (see below).
This web page (made by an anonymous person who implemented a perl program from a well-known book) generates random text based on text you give it, using a Markov Chain algorithm. Put simply, it analyzes the text you give it to see what words most often follow what other words, and then uses the probabilities to randomly choose words, starting with the first word of your input text.
The “size of word groupings” parameter on the form tells it whether to look at just the next word in your input text (if you enter “1”) or the next two words, etc. The bigger a number you enter, the more the generated text resembles the input text, so that a value of four (the maximum) gives output with big chunks of your original text.
The bigger your input text, the better and more varied the result will be. You get different output each time you generate text even with the same input because of the randomization.
