Poetry: language aspiring to a machine-like state

Andrew Joscelyne
3 min readFeb 9, 2021


Poetry is language in free fall — no one is there to be held to account for its meaning. This is why machine verse can sound so convincing: it can be synthesized from anything in the canon, and challenges us to use our imaginations to make sense of it. Just like all poetry.

The deeper interest of the neural AI program Deep-speare, then, lies in what is beyond the fact that any number of poems, sayings, proverbs, and formulaic language entities can be generated from data stashes by tweaking the weightings inside an algorithm. These outputs will be synthetic text that is destined to mingle into the human canon as time goes on, quietly changing the percentages of human and machine contributions to discourse in general.

AI is indeed a technology that augments — in this case the human art of sonnet-making. The reason why this algorithm seems so successful is that, for centuries, the language in which human poetry has been written has been trying to approach the haeccitas of artificial language generation! Whether by means of rhyme, rhythm or numerical calculus over a fixed set of syllables, poetic language has taken good old human language and used it to perform highly mechanical speech acts!

Poems can be created out of almost any word stuff. They emerge as pure performance, rather than part of a conversation around the world’s business. They don’t really do the things humans standardly do with words, such as questioning someone, explaining something, or laying down the law. Or rather, they only do them as if… Poems are the epitome of language games. Their aim seems simply to summon emotion, not respond to it.

There is therefore nothing more machine-like than verse: you can generate a poem by cutting up a page full of any old written sentences into a confetti of words and then rearrange them into a poem on the scaffolding of an initial grammatical sequence. No one can tell you it’s not poetry…

Exquisite corpses, silly rhymes, Augustan periods, onomatopoeic echoes — all these tricks can be used to collapse everyday speech patterns and typical narrative styles into small time-bombs of language. Which is why a machine could easily be programmed, had it but data enough and GPU power, to churn out multitudes of similar texts. It is human appreciation that says it sounds like plausible verse, and in the last resort almost any juxtaposition of words can be used to evoke something — raw or cooked, ancient or modern.

Poetry generation indeed was one of the first extra-curricular uses to which access to a computer was put by various groups starting in the early 1950s. Largely no doubt because poems can typically be short but unusual, and therefore not require too much expensive data or processing time.

In 1950s Italy, poets such as Nino Balestrino dabbled in digital verse, as did the French Alamo group in the early 1980s, as an offshoot of the playfully formalist OULIPO movement. Earliest of all in 1952, Christopher Strachey, an English computer pioneer, wrote a text-generating program whose sole purpose was to produce love letters, much aided by a perusal of Roget’s Thesaurus. A close cousin, then, of love poems…

A non-computer program with similar aims was also proposed by the French writer, polymath and OULIPO founder Raymond (Zazie dans le métro) Queneau, who published what we could facetiously call a remarkable “poemutation” — Cents mille milliards de poèmes — in 1961. Ten sonnets were printed out with each line on a separate strip of card. They all use the same rhymes and schema so that any lines from one sonnet can combine with any of the nine others generating the 10 to the 14th different poems referred to in the title.

Just as we are bound to generate synthetic forms of life by tweaking the grammars of genomes, so we shall be able to play with the grammars of poetry and then prose perhaps, adapting our languages to wild new prospects of expressiveness. But remember that the machines we know today at least have no idea what they are doing or why. They might stop producing poetry if they did…