31   Allison Parrish Sends Probes into Semantic Space

[Computers] are really good at creating texts that don’t work in the way that people expect language to work.

—Allison Parrish78

“I’m an experimental computer poet,” says Allison Parrish. “I write computer programs that write poems. It’s an awesome thing to be able to put on your business cards.”79

Parrish is a one-off, full of forceful opinions, humorously expressed. At the University of California in Berkeley, she studied linguistics and French. Before that, she completed a degree in computer science, working her way through university as a computer programmer. She has been programming since the age of five. Parrish went on to complete a master’s degree at NYU’s Interactive Telecommunications Program (ITP), where she is now a professor. She describes herself as a computer programmer, poet, and game designer who is fascinated by the fact that “computers are really good at creating texts that don’t work in the way that people expect language to work.”80

To Parrish, code is a tool that opens new avenues for poets, in the same way that abstract expressionist painters Mark Rothko and Jackson Pollock used the tools of canvas and paint to subvert what a painting can do. For poets, code offers new and expressive ways to manipulate language.

“I always seize authorship for myself,” says Parrish. “When I put out a book of poems it’s by Allison Parrish, not Allison Parrish and a poetry bot … in the same way that a Jackson Pollock painting is not by Jackson Pollock and a paint can.”81

To put it another way, she disapproves of emphasizing process at the expense of product. When she writes text and poetry, it’s with a reader in mind. She believes that there should be a definite connection between intention and output. She makes decisions about each word, about the range of vocabulary she draws from and the algorithm she uses. “These are all incredibly expressive decisions.”82

Parrish also disapproves of what she sees as a goal of computational creativity: to tune algorithms so that they produce work as close as possible to that of human poets. “I’ve made it my goal as a computer poet not to imitate poetry but to find new ways for poetry to exist.”83 The computational poetry of the future will, I suggest, be something we cannot even imagine today. Parrish points out that we’ve been doing computer-augmented writing ever since the introduction of the word processor. We use autocorrect, spell check, and the machine’s thesaurus, while the keyboards on our phones continually guess our next word.

Much of computer-generated poetry is gimmicky, Parrish feels, but it gets a great deal of press for the simple reason that it’s generated by AI. She looks forward to a future in which there will be more interfaces explicitly for writers, driven by AI and machine learning. She expects the effects on the style and rhetoric of written works to be extremely subtle and sophisticated.

“A text is poetic,” she says, “to the extent that it is drawing attention to the affect that it has above conventional meaning.”84 She draws a parallel with the invention of photography, which freed artists from having to depict the world around them realistically and enabled figures like Claude Monet to emerge to create more impressionistic images of real life. “That is the potential of computer-generated language,” Parrish says. Language can spark the brain in the same way that visual images do. “How can those parts of the brain be made to tingle in a way they have not tingled before?” she muses.

Modernist poetry, she says, arrived with the invention of the typewriter, which offered a way to produce not only imaginative text but imaginative image. The digital format is highly suitable for poetry in that it offers an elastic and quick way to compose in a variety of computer languages.

How do aesthetics and beauty apply to coding? In coding, Parrish says, she aims for elegance in procedure. “If a procedure is elegant, I mean that somebody can guess the way the procedure works from the way the output looks; that to me is really important.”85 She compares this with conventional poetry, in which the reader’s understanding of the rules of the poetic form used—the sonnet, for example—makes the poem more enjoyable to read. She is fascinated by the rules that go into the making of a poem, whether it be conventional or computer-generated.

Says Parrish, “Poetry is much more interesting if there’s a preface that explains what the poet was thinking. … For me it is more interesting to read books about poetry than to read poetry.”86

Parrish’s great inspiration is a quote from Jean Lescure, a member of the avant-garde French writers’ group Oulipo, founded in 1960; the group created works that were the predecessors of today’s conceptual poetry. Oulipo included luminaries such as George Perec and Italo Calvino. The words that struck Parrish’s imagination were, “Like mathematics, literature could be explored.”87 Parrish likes to think of herself as an explorer, not of outer but semantic space, the space of language.

A lot of her poetry takes the form of Twitter bots. She considers this the ideal way to explore semantic space, akin to exploring space with space probes, which are essentially robots. Her “favorite space probe” is Voyager 2, which explores the outermost planets and heads ever deeper and deeper into space, entering areas utterly inhospitable to human survival. It has been on its mission now for over forty years. While we live safely ensconced in our own familiar environment, space probes venture into the unknown, sending back reports using radio signals, and often do not return, which indeed will be the fate of Voyager 2.

To explore semantic space, Parrish came up with what she calls a very simple literal bot, “a robot that deals with words and letters.”88 She uses a Python program that reads all the lines from a given text, inserts them into a data structure, “then spits them back out in random order.” Just as Voyager 2 explores a part of the universe where people have never been, Parrish’s “literal bots” explore inhospitable parts of semantic space, beyond the kinds of language that we know, currently designated as nonsense. She likes to quote Julius Caesar’s words in Gallic Wars: “It was not certain that Britain existed until I went there.”89 Similarly, she explains somewhat whimsically, she decided to visit the unknown areas of semantic space with her literal bot to find out what was there—as she puts it, “exploring (semantic) space with ‘(literal) bots’ because humans abhor nonsense (and need help finding a path through it).”90

So what comes out of all this theory? Parrish’s first venture into semantic space uses n-grams, sequences of a fixed number of words. Two words make a bigram, three words a trigram, and so on. For data, she uses the Google Books Corpus, consisting of many millions of words. From this she assembled bigrams in the form about, anything, 124451, meaning that the word about is followed by the word anything 124451 times in Google Books. To keep the exploration within limits, she used only bigrams in which both words begin with a. She laid out the words in a matrix beginning with a, horizontally and vertically. Where there were no cases in which the two words followed each other, she left a blank. Parrish then used an algorithm that created a visualization of these sequences in two-dimensional space, with larger rectangles where there are more occurrences of a particular bigram, smaller where there are less. The blank areas that mark the absence of word sequences is empty space—the void.

When she did the same thing with trigrams, it produced a rather spectacular three-dimensional visualization with words and cubes slowly revolving in space, “like a scene from some weird space movie.”91

The empty spaces—the great voids—occur where there are no instances at all of a certain word following another. Parrish picked out some of these nonbigrams:

All words that have never before been used together, at least in Google’s Book Corpus. These are juxtapositions that “have never been thought of or explored before, just because of how we conventionally think about the distribution of language.”92

Parrish compares the difficulty of engaging with nonsense with the difficulty of surviving in outer space. She likes to think of herself as a proud descendent of the Dadaists, such as Tristan Tzara, one of the first to take language soundings—like taking soundings using radio signals from deep in the ocean or high in the atmosphere. In 1920, Tzara famously wrote instructions for how to compose a poem: cut up a newspaper article into individual words, jumble them up, then paste them onto a sheet of paper. Voilà, you have a poem.

Parrish also calls her “venture into nonsense” a sounding—sending a probe to find out what’s out there. In 1920, the process had to be done manually. But all the same it was akin to an early computer program. Then, in 1959, Theo Lutz created the first computerized poetry generator, which spewed out sentences with the words chosen at random, nonsense sentences. For Parrish, this was the first literal robot exploring semantic space.93

At first glance, says Parrish, the Twitter reports tweeted back from Voyager 2, more than 16 billion kilometers away, seem like the work of a procedural poet:

FAQ How fast are you going? Not as fast as you might think, but still at a good clip! Sister V1 [sister ship Voyager 1] is leaving the Solar System at ~61,391 km/hr, and I am leaving about 10% more slowly, at #55,644 km/hr.

Voyager 2 has tweeted almost sixteen thousand times, and more than 350,000 people follow it.

Parrish’s Twitterbot @the_ephemerides juxtaposes NASA images from space probes with words chosen at random from two large bodies of text—one on geology, the other astrology. These have immediate relevance to photographs of planets, making it easier for readers to see into them meanings that could be construed to have depth. Thus, alongside a photograph of Saturn’s rings is this poem:

We cannot get away

from it. The ocean are by

no probably does not

likely as people proceed.

Anyone who has breathed depth into fragments of Beat poetry can do so here.

Parrish sees her most famous Twitter bots as a form of computer-generated poetry. The best known is everyword. Starting in 2007, it tweeted every word in the English language in alphabetical order, one every half hour. At its peak, it had over one hundred thousand followers. If you were to enter the everyword words into Parrish’s two-dimensional bigram matrix, described earlier, they would form a straight line cutting diagonally across the matrix, because each word only occurs once and in juxtaposition to the one that precedes it alphabetically. The resulting line is a bit like the trajectory of a space probe as it sends back its reports, says Parrish.

Tony Veale, whom we will discuss in chapter 34, says of Parrish’s Twitterbots: “Her bots are simple from the AI perspective but ingenious from the artistic perspective.”94 He too compares them to the cut-up techniques of Beat poets like William Burroughs and Brian Gyson. They are “more sophisticated versions of those techniques,” he says.

Another venture into the realm of nonsense is Parrish’s Twitterbot PowerVocabTweet, which makes up new words and definitions. It generates new words by splicing together two existing ones on the basis of the number of letters and syllables, and creates new definitions using a Markov model based on word definitions from WordNet. The result is a stream of words that we could never dream up but which surely fill gaps in semantic space.

@PowerVocabTweet has over three thousand followers and has tweeted over twenty-two thousand times. It generates words and definitions such as the following:

  • aghbridge, n. (football) the running back who plays the alto saxophone
  • durotic, n. a common bean plant grown for its beautiful song
  • eight-billed, adj. great in degree or intensity or amount

PowerVocabTweet boldly proclaims, “Boost your vocabulary with these fiercely plausible words and definitions.”95 Nonsense though it is, Parrish speculates, there may be some people who really think they’re increasing their vocabulary by following it.

Parrish considers her work to be calling for an “ethics of semantic exploration.”96 She wants “to remove the value judgment between stuff that makes sense and doesn’t make sense—to point out the fact that, to some extent, that distinction is arbitrary. … In fact it’s the mission of poetry to show how arbitrary that distinction is.”97 She is entranced with the way that computers can freely create nonsense—that is, “texts that don’t work in the way that people expect language to work.”

To return to Caesar’s statement in Gallic Wars—“I was not certain that Britain existed until I went there”—Parrish suggests that “perhaps unexplored parts of semantic space have been previously visited,” in the same way that Caesar was wrong; there already was life in Britain long before he came, saw, and conquered.98

Just as exploration and discovery all too often end in the exploitation of the newly discovered country, so too maybe the gaps in semantic space are there for a reason and we have no right to go there. And if we do, we cannot assume we have the right to appropriate what is there. An example is the language of a newly found group of people. Similarly, Parrish abhors appropriating the text of others and repudiates Kenneth Goldsmith’s notion of “uncreative” writing involving the “copying, recycling, or appropriation” of other people’s work.

“I’m going to be a hardliner and say that computers cannot be creative,” Parrish tells me, insisting that there will always be a human hand behind them.99 “Furthermore, it will always be a mistake to attribute volition to the computer and not to the people who programmed it because attribution of volition is removing personal responsibility: the algorithm did it, not me.”

In her recent book, Articulations, Parrish ventures into the world of sound.100 She’s chosen two million lines of poetry, from which her computer programs seek out words and lines that echo each other sonically to create poetry in which sound rather than meaning is uppermost. The resulting poems are “a lot of fun to read out loud,” she says.101 Here is the beginning of one:

And Then She Went Away. She Went Away, Away, She Went Away. She Went Away

Thunder, lightning, fire and rain, and laughter, and inn-fires. After the fire of London and another of a pastoral vein—the venerable original the adolescent and the venerable, richly the upland and vale adorn, Buddha, the holy and the benevolent, of many a lover, who the heaven would think …102

Parrish insists that computer programs will not replace poets but will take on exploratory work that humans would rather not do. She continues, “Because a computer program isn’t constrained by convention, it can speak interesting truths that people find difficult to say, and it can come up with serendipitous juxtapositions that make language dance and sing and do unexpected things that can be beautiful and insightful.” And perhaps creative, too.

Notes