Reading poems made by maths

A while back I wrote about playing around with tweets, Markov chains and the shipping forecast to make little audio experiments. This is the second part of that. Probably not much point reading on if you can’t listen to the audio clips below, so get some headphones and set a few minutes aside…

Brian in the booth

The Shipping Forecast has become a poetic touchstone for the national subconscious. The forecast itself is a person reading a list of words and numbers generated by equations modelling the weather.

I wanted to see what it was like to professionally record something made by maths.

I took the text of Songs of Innocence and Experience and once again plugged it into Cheney – a Markov text generator built as a web-application by Tom Armitage.

The app takes a word (or string of words) and works out the next likely word, doing the same to that newly selected word and so on and so on. In this case it does it enough times to spit out some text that mathematically represents a Blake-like poem. Something like;

The kingly lion old folk

Thou, mother weep!
When the child Weep,
nor birch

Where are wasted in morning,
Joy I must love delight,
Where the summer’s play
Round the fire of winter there

And, father, rejoicing to
Thought best
To welcome in learning’s bower,
With hollow piteous shriek

I took a selection of these poems into a studio and worked with actor Brian Fenton and producer/director Ann Scantlebury to see what happened. Would meaning emerge?

The poetry of the shipping forecast is a mush of the speaking, the structure and the intellectual distance (if you’re listening on Radio 4 tucked up in your cottage then it’s not really for you – poetry comes from the privilege of your life not being in danger).

These mathematically generated poems (and they are poems) don’t have linear interpretations. They are, to a certain degree, nonsense. But it’s the kind of nonsense that leaves room for character. Brian had to judge where to take these, what tone emerged from the text.

The results sound older than Blake’s verses. Blake’s affectations become uprooted by maths, compressed by probability into strings like “As thy softest care/Met in a heaven the maid“. The recordings were things that seemed to echo across a few centuries and roll around the booth.

Some, of course, sounded like straight-up lifts from the original. Twiddling the depth of the Markov chains did interesting things, and less interesting things.

Sometimes it was a struggle. The words don’t feel like they fit, which meant it got a little bit Toast of London a few times (“Have fun with it”), but Brian and Ann did an excellent job and I really like the results.

I also know what it’s like to be in a studio now. And that’s probably the most important bit; part one was about exploring random text generation for reading, this was about professional recording and the context and process involved in that.

I had it easy. Brian and Ann were great and extremely patient with what I was super-worried was an odd damn thing. But Brian has to sight-read things every day, often with very little preparation and with far less context than I provided. His job is to make meaning emerge. As a director, Ann’s role is to behave as first listener (Is the meaning legible? Is it internally coherent? How can I improve the delivery to move the listener?) and make the recording sound sharp. The provenance of the text plays a part in that, but more important is what happens when it’s actually spoken.

So, that’s the end of part two. The next part gets back to the question that started all this, a question I might well be apocryphally attributing to Warren Ellis; ‘What is a shipping forecast for the internet?’

That, I think, requires a research phase before another making phase. How was the shipping forecast made and formatted? How did the recording process solidify? Where are the places that predict behaviours of the internet? What will I define as ‘sectors’?

Huge thanks to Ann and Brian!