New Minstrelsy of the Scottish Border

The Computer Writes a New Old Language

There is an idea about poetry, developed relatively recently in the human story, that it is the result of a mind making something new: a mind expressing itself, channelling inspiration into original creation, interacting with the world as a sovereign entity exerting its aesthetic will, being an author. This idea of authorship tends to mislead us when we think about computers making poetry. When we see a poem that has been, as we say, “written by a computer”, we immediately imagine an advanced thinking engine, possibly in a chic humanoid shell, churning its many layers of numbers to think thoughts beyond human understanding, authoring art to rival our own, inevitably rising to destroy/master/transcend the limitations of flesh, &c.

But there is an older idea of poetry – of art – which can maybe take us somewhere more useful, for now: that poetry is the craft of combining inherited and accepted elements in exciting ways, displaying the poet’s skill in memory and recombination, with new flourishes and touches of original language introduced rarely and judiciously, only when they enhance the poet’s and audience’s connection to the deep, historic whole. This approach to poetry is why the sagas reproduce the same kennings over and over, why the Odyssey’s dawn is always rosy-fingered, why the battles in the border ballads seem curiously alike even when conducted by different characters in different wars. This is also why alliteration, assonance, rhythm and rhyme exist (have you ever tried to memorise an epic without them?)

This idea of art as interesting recombination rather than original creation helps us to understand computer poetry better. What our poetic machines do now is to combine and recombine the elements of language, with programmers to push them to do it in more interesting ways and editors to help weed out the duller or clunkier results. Computers can write poetry in much the same way as poets have always done, by taking the poetry that already exists, making sense of how it works, breaking it into its elements and putting it back together in interesting ways.

But creating this binary system of recombination and creation, craft and art, natural and unnatural, risks, as all such systems do, ignoring the ways the two halves are always making and collapsing into each other, especially when the binary is given a pseudo-historical linear development, as with this introduction.

There is, after all, no creation that isn’t an extremely complicated act of recombination, and no recombination that doesn’t require some kind of authorial creation. And so to think hard about this history of poetry is also to think hard about poetry’s material: language.

The poets of minority languages, especially those minority languages like Scots which are subject to progressive erasure – linguaphagy – by a larger and more successful linguistic neighbour, embrace a variety of tactics to recover their tongues. There is retraining your ears to hear what is actually spoken by your friends, rather than what the dominant orthography thinks is spoken; there is mining the literature and the lexicons for words that are nearly lost; there is creating new and old-new words to express what would have been going to be expressed had the languages taken a different path through history. There is, in deploying these tactics, a negotiation between the poet and the audience’s almost-understanding so that communication takes place. As the words unroll, it’s never entirely clear whether anyone really understands what they really mean, or rather, whether the conditions currently exist to realise the potential for meaning within them – the potential which has been lost, which is being recovered, and which is imagined into the future by the very act of speaking.

All of which is to say that the understanding of Scots as a “synthetic” language is to make the same peculiar and politically-loaded binary assumptions about language as we have become accustomed to make about art: that there is either natural language that evolves by itself through recombinatorial development or synthetic language that’s created by an author sui generis and ex nihilo. But, of course, linguistic change was always shaped by poets’ creativity, and poets’ new languages always had deeper foundations in the vernacular and historic than they or others would care to admit.

When a computer program attempts to write in a minority language, it appears gloriously shod of all the political biases of the poets making the same attempts: the computer does not “understand” English any more than it “understands” Scots: it only knows the patterns of symbols and what patterns are likely to follow those patterns. But programs are written by people, and when, as in this case, one is using systems developed by the American military to deduce the pronunciation of written text, the politicised results are visible in the jankiness of its rhyme schemes, and when the computer has been tuned to match a corpus that cannot decide how to spell a Scots word because it was written at a time of crucial transition between Scots and English as the local lingua franca, the ballads frequently wander down orthographical cul-de-sacs which even the program’s authors cannot identify as either Scots or otherwise.

This is what we’ve done: we met in the contested border town of Berwick-Upon-Tweed, built a computer program from repurposed old code and self-authored new code, tweaked and tuned it to best fit the entire ballad text of Walter Scott’s Minstrelsy of the Scottish Border, and set it to write some new ballads in the same form and language. In doing so, we are asking of it no more than was expected of the ballads’ first singers; in recording the results, we are as partial and editorial as Scott. We hope that the program’s many successes and many failures, and in particular the messy border zones between success and failure, teach us something about language, something about poetry, and everything about what it means to create.