the preoccupations of habitual excessively fond
she had human tendencies and
wished to all the last words
his presence of anything
there is nothing to the preoccupations of habitual
excessively fond of the few who herself
but this beyond a distance and as
turning with more than a meditative turn
that had been sweeter by warbling at her eyes a sudden
he felt her head including face of listening
who his feeling as he had
a horse has the fellow of various
and he he has perhaps in a prospect of
womanhood the world publish his pain she had early the burning
being something else be it came wet wet wet
but I dare love that the beginning
August 8-9, method 770ac8b0-5626-426c-bb79-1adf9ad13324 (human accomodating increasing levels of machine randomness), Text: middle sutra manifesto (Middlemarch by George Eliot, the Kama Sutra of Vatsyayana, the Manifesto of the Communist Party by Marx and Engels), generator: epogees
yo I just got my hands on a copy of the Oulipo Compendium. Killer book of all time. I dig the way they come up with authoring algorithms. Only I don’t like the way they name them. I think they should be UUIDs or something, with a parenthetical nickname.
Anyways, one thing I realized while testing out all of the authoring tools I was looking at recently is that the process that the human author uses is important. (i.e. what decisions is the human making in the algorithm? once the lines are generated, how are they changed? are they re-generated as in Gnoetry, are words from the language model added as in epogees, are lines selected as in mchain/epogees?) for me, Oulipo highlights the importance of those kinds of decisions. Anyway, in that spirit I decided to formalize one of the authoring styles I found myself using in epogees, and which was used for the poem above.
Method 770ac8b0-5626-426c-bb79-1adf9ad13324 (human accomodating increasing levels of machine randomness) Summary: the human uses a generator to select and arrange meaningful text fragments. the human then generates a number of lines, and uses them (with limited editing) to complete the poem. A. The human creates the meaningful core of a poem, using computer-generated text 1. generate several (4-14) lines of text using an n-gram generator (such as epogees, gnoetry, janusnode, processing/RiTa, etc.) a. suggestion: generate text with high levels of alliteration or assonance if possible. if not possible, use only 1-2 source texts. this is for coherence. b. suggestion: each line should be fairly long (8-9 words) if possible. this is so when you remove words in 2.a, you have something left that might be interesting. 2. pick interesting phrases from that text, to create new lines a. suggestion: for each line generated in A.1, remove the words that are not interesting. see what you have left. b. suggested constraint: you may only remove words from the beginning or from the end of a line. if you remove words from the middle of a line, you must break the line into two lines. c. suggested constraint: you may break a line in two, but you may not combine lines. 3. see if these any of these lines can be re-arranged to create a "core" of more overtly meaningful text. if they cannot, return to A.1 and generate again. a. variation: you keep a "pool" of lines created in A.2. throughout the composition of the poem, and can use them at any time b. variation: you keep a "pool" of lines created in A.2, but you erase it when you move to step B. c. variation: you do not keep such a "pool". instead, every time you generate in A.1, you erase any unused lines. d. variation: every time you generate in A.1, you erase ALL lines, even those you think would make a good core of a poem. e. variation: the "core" is only at the beginning or end of the poem f. variation: in step A the "core" is made up of 2 stanzas (of 2/4/6 lines) g. variation: you may add words that are in the n-gram model, using something like the "suggest next words" feature in epogees. B. The computer generates several lines of code, at least 75% of which must be used by the human 1. generate several lines of text using an n-gram generator a. suggestion: generate relatively small lines (because you won't be removing any words, only entire lines.) 2. for each 4 lines generated in B.1, you may remove 1 line. (So if you generated 5-8 lines you can remove 2; with 9-12 you can remove 3, etc.) a. variation: the final number of lines generated in B1 that remain after step B.2 must be equal to the number of lines generated in step A. b. variation: you may only remove 1 for every 4 *adjacent* lines. so if you generate 12 lines you cannot remove 3 lines only from the end of the set. the lines you remove must be more evenly distributed. c. variation: you must remove exactly 1 line. otherwise, you could choose not to remove any if you didn't want. 3. The lines remaning after step B.2 are placed among the lines of poem generated in step A. a. variation: the lines from B.2 must be kept together as a stanza (or as 2, 3, or more stanzas) b. variation: the lines from B.2 must be placed between the stanzas from step A (which are themselves placed at the beginning and end of the poem as per step A.3.e)
The goal is to produce a text that balances coherence and randomness to produce a text open to a variety of meaningful interpretations on the part of the reader.
btw I think the human parts of this method could be automated. They’d have high error rates, of course, but poetry is a forgiving domain. (I don’t think I’ll do it though.)