In this video, Vlad Butucea introduces us to an algorithm that he has trained to generate Edwin Morgan-like poetry and discusses translation, queering language, dominant languages and the human/ non human. The video and full script can be viewed below.
Edwin Morgan Bot. The Computer’s Other Realities
The whole life is adornment of murning in its edge.
What is a thousand poems like lumber.
What is nothing like the moon to change we were ready.
This is a poem written by a computer. I trained an algorithm to ‘learn’ the complete works of Edwin Morgan, and now it can write its own strange little poems. Recycling and re-imagining Morgan’s words. Here’s another example:
The streets are strangers,
The Streets are stretched on the streets.
The streets are strangers and stars.
The streets are stronger, and the streets are strangers
In the streets of the streets of the streets.
As you can see, a computer may not become Makar anytime soon, but I think there is something in equal measure endearing and thought provoking about computer poetry. My name is Vlad Butucea, and in this video I will conjure the inspiration of Edwin Morgan, and the dreamy artistry of recurrent neural networks to think about language, poetry, and the politics of decentring human creativity through technology.
But before I get into that I wanted to say more about how I arrived at this project. It all started with a Second Life Grant from the Edwin Morgan Trust which gave me the possibility to become immersed in Edwin Morgan’s work, and to discover a poet with whom I was completely unfamiliar. I was given the time to read and to explore and to get myself lost in Edwin Morgan’s universe. Early on this journey of exploration, I cam across a poem called Copernicus from his epic Planet Wave (1997). And there was a line in there that really inspired me: “no stories are told about this man who kicked the Earth from its false throne”.
The man Morgan writes about is Nicolaus Copernicus. In the fifteen hundreds, he created an earthquake, a seismic shift in how humanity saw itself. He described a universe in which not the Earth, but the Sun, was at the centre. Indeed, he kicked the Earth from its false throne. Copernicus made the world aware that in the vast and unknown Cosmos, there is little exceptional about the human. A difficult realisation. One that, to this day, I think, we are coming to terms with.
‘No stories are told about this man who kicked the earth from its false throne’.
And I began to see Edwin Morgan’s work in a new light. As a poetry of decentring. A poetry that kicks things off false thrones and gives voice to otherness. Monsters, aliens, animals, computers – are all equal agents in Morgan’s universe. They speak, they act, they think. And, by decentring the human, and by giving a voice and a language to the non-human, Morgan engages in a deeply political act.
For example, when The First Men on Mercury exclaim ‘Bawr stretter! Bawr. Bawr’, they refuse to
engage with the language of their human colonizers . When the Loch Ness Monster sings ‘Gdroblboblhobngbl gbl gl g g g g glbgl’, it voices knowledge unknowable to its human hunters.
When the computer fails to wish us a
jollyberry’ Christmas, it shows the impossibility of human language to contain all communication. Morgan tells us there is more to life and to the world, than a narrow, human-centric imaginary. He exposes a diversity and a vibrancy beyond it. He challenges human-centricism, and I think, in wider terms, he also prompts us to question and to challenge other centres, other false and unearned thrones: The West. Masculinity. The Empire. Heteropatriarchy. Whiteness.
As a writer, I find Edwin Morgan’s games of language particularly inspiring. If English is the dominant language of representation, a centre, then Morgan’s many non-human languages trouble it. Reflecting on the underwater otherness of the Loch Ness Monster Song, Dr Monika Kocot argues that ‘the poem deconstructs two myths, the myth of superiority of English and that of conventional sense creation’ (2016, p. 119).
Morgan’s famous computer poems do something similar. They decentre English and highlight alternative ways of making sense. In the Computer’s first Christmas card, we have quite a funny and endearing computer that tries to emulate English, but ultimately fails at the task. In the Computer’s second Christmas card, the computer begins to move beyond English. A few words: kinds, feast, scrub, carol – remain. But most other words are unknown to us. In Computer’s first code poem, the machine has acquired a distinct language altogether, unintelligible through the codes of English. The computer speaks in its own language. These poems, and others like them, undermine the centrality of English as the dominant meaning making system.
TEYZA PRQTP ZSNSX
The thing is though, Edwin Morgan never used a computer. He imagined languages for computers, he gave them a poetic voice. But never used one himself. So, with this project, I decided to take Morgan’s computer poems to a more literal conclusion. What happens if a computer was to actually study Morgan’s work and then write its own poetry?
So I used a recurrent neural network algorithm – created by Andrej Karpathy – and trained it to ‘learn’ a big chunk of Edwin Morgan’s work: the most recent Centenary Collected Poems, Beowolf, his Translations and many more.
The code is a ‘multi-layer Recurrent Neural Network for training and sampling from character-level language models’ (Karpathy, 2016). In other words, when the algorithm is fed a sample text, it studies it again and again until it can generate its own text based on what it’s learned. The more it trains, the more confident it becomes, until it can create its own words, sentences and even full poems.
That said, the neural network does not simply reproduce the sample text – instead, it breaks it apart, analyses it, and tries to understand it through its own ways, through its own mathematical frameworks. This particular neural network studies the probability distribution of a character in a sequence. So, for example, if given the sequence S – T – A, it may be able to predict that the next letter is most likely an R. STAR. It will then continue to predict the next character producing further words and even sentences.
This is fascinating to me, because it’s a unique way of creating words, that’s very different from how we create words. The sequence of letters S T A R in English is not determined by probability – instead, it is purely random. We use this random sequence of letters to describe a celestial object. But we may as well have used this sequence: SATR. Or this sequence: RATS.
However, as far as the algorithm is concerned, there is a statistical reason why R should follow from A and T and S. It interpreted our language through a mathematical framework, and is now able to emulate it, with some degree of intuition and success. But, even if it arrives at the written word STAR, it has no concept of ‘star’ as a celestial object – as we do. For it, STAR is a probabilistic result. It’s maths. It’s numbers. It’s STAR and our STAR are two different things, two different concepts, two different languages underpinned by different rules and logical frameworks. The bot can write in what may look like English, but the ways it understands and uses language are different. It appropriated English and gave it its own spin.
So let’s look at some examples. This is a poem generated from a low training state, where the algorithm is still in its learning infancy. As you can see, it is pretty much nonsensical as far as English is concerned:
I hemoe mive is a streem Ghellist me:
rlenging to deflears on year no fark gook clester in heart
There are some words, and what appear to be sentences. But there is little I can actually understand from it. This reminds me a lot of The Loch Ness Monster song, a language unintelligible to me, the human seeker.
In this other example, it was able to learn the word ‘the’. And reproduces it frantically and quite amusingly:
e the the the the the the the the the the the the the the the the the the the store of the the the the
the the the the the the the the the the the the the the the the the the the stand the the the the the
the the the the the the the the the
But here’s what happens after a longer training session. As you can see, the poems become more complex, and we start to recognize more English-looking words. There is also some basic use of grammar – there is subject and predicate – and even a verse with distinct lines.
and the stars are still the stars and the stars and the stars
and the stars are all the shadows of the stars.
The streets are the stars of the stars
And finally, this is the most complex training I was able to achieve, after about a week of trial and error exploring different parameters to create the most comprehensive model I could. This is supposed to emulate the sample text most confidently, to be the most accomplished Morgan bot.
Gods have something on the sheep of the streets,
On the skin of its strike the heart,
The sound of men and the green stars,
In this state, the bot is not only able to predict words. It can make sentences with quite good English syntax and even punctuation. It uses lines, and I think it is quite rhythmical, as Morgan’s poems often are. But, even if this appears to be English, there is something odd and otherworldy about it. It is English but also not quite English. English that does not make conventional sense. This is because the algorithm does not produce meaning in the same way that human language does. It’s rules are mathematical, rather than driven by human-like reasoning, or logical associations. This is why, even if it can emulate words, syntax, or rhythms, it does not make human-like sense. It has it’s own way of making sense, it’s own rules to language.
While some may see a future where computers are able to speak and write like humans, I am far more interested in computers writing and speaking like computers. Writing with their own agency and rules, that don’t follow from human language.
And I think this is something that Edwin Morgan was trying to achieve with his own computer and other non-human poems. To invite us to accept that there are other ways of making sense beyond our own language, beyond our own sense and understanding.
In 1968, Morgan contributed to the Cybernetic Serendipity conference in London. This was a major exhibition, a hugely ambitious project aiming to explore the burgeoning role of computers in art. It was here that some of the first computer-generated poems were shown, the so-called computerized haikus. Reflecting upon this major development, Morgan said:
Although I am interested in real con1puter poetry (and art and music), my special concern in these poems has been to take an ironic but not antipathetic look at the relations that will exist between computer creativity and human creativity, the challenge to the second from the first, the probability of a new approach to at least some aspects of poetry, even a deliberate emulation of the so-called blunders or digressions which at times arise (one would say) creatively within a computer context (cited in Reichardt, ed. 1968).
I was inspired by Morgan’s invitation to challenge human creativity through computer creativity. I think the Morgan bot disturbs, decentres, and takes away some agency from Morgan the poet. It appropriates Morgan’s canon and turns it into its own, algorithmic, weird art. Developing this project, I realized that the question is not how to get bots to write poetry exactly like Edwin Morgan. But to enjoy how bots write poems on their own terms. Through their own rules, logic, language that may make little sense to us. To see poetry as an art beyond the human, playful and absurd, challenging and unpredictable. To shift poetry, to shift words, to shift language away from its human centre.
When Edwin Morgan invents computer languages, alien languages, monster languages he creates, as Professor Chris Jones argues, “a democratic imperative towards universal polyphony, a sense that no one voice or attitude ought to be privileged above all others” (2006, p.lxvii). By creating this universal polyphony, Morgan ultimately calls for better relations between us all – human or not. He calls for a world of no false thrones. A world of no dominant centres. And ultimately, I think, this is a vision of a better world.
PS: There was no way for me to fit this in the main body of the video, but I really felt the urge to share this with you. This is a result I kept getting. Across multiple training session, across different data sets and parameters, I noticed that the algorithm always ended up obsessing over the word: ‘star’. Here is one of my favourite poems it created, it’s called pery stars.
The starless the streams and the streets and starsed the strange of the stone.
The starless the world of the strange the stars and the stars.
The sun was a shiner of the stars and stars and shouts
I don’t fully understand why or how this obsession with stars came about. I suspect it has to do with statistics: words with an S are more common, and often an S is followed by a consonant. But that still does not fully explain why most of my training sessions resulted in a star-struck, dreamy algorithm – the word star only appears about 80 times in the dataset and definitely is not the most commonly used word in Morgan’s writing. I want to believe that, at some level, the computer found the Cosmos in Morgan’s writing, making some connection to him all the way from Glasgow to Saturn. And I became quite attached to the Morgan bot because of this.
Jones, C (2006) Strange Likeness: The Use of Old English in Twentieth-Century Poetry. Oxford: Oxford
Kocot, M (2016) Playing Games of Sense in Edwin Morgan’s Writing. Frankfurt: Peter Lang.
Murray, N (2020) Edwin Morgan’s Computer – Interactive Games, Edwin Morgan Trust, National
Reichardt, J. ed. (1968) Cybernetic Serendipity. The computer and the arts. London: Studio
char-rnn, Created by Andrej Karpathy: https://github.com/karpathy/char-rnn
Used under MIT Licence.
air-drops-master by John Bartmann
Used under licence CC0 1.0 Universal (CC0 1.0) Public Domain Dedication
This project was funded through The Second Life grants scheme, celebrating the Edwin Morgan
Centenary. Supported by Creative Scotland, The Saltier Society and The Edwin Morgan Trust.