When neural nets do carols: 'Santa baby bore sweet Jesus Christ. Fa la la la la la, la la la la'

A truly AI Christmas to one and all + El Regmas song

What is this latest madness? you may ask. Well, it’s 2017 and it’s Christmas so someone has obviously had a crack a training a neural network to write carols.

The results are at least entertaining, even if they might offend the devout – or simply those not keen on the idea of Santa popping out a baby Jesus or people getting it on with snowmen.

Janelle Shane, a PhD student from the University of California San Diego who likes to use neural networks for humour, is responsible for the festive frivolities.

“My process starts with a dataset – something that the neural network has to figure out how to imitate,” she wrote on her blog.

“Rather unfairly, I give it no instructions about whether it’s trying to write knock-knock jokes or invent Halloween costumes or begin a novel. It doesn’t even have any built-in knowledge of English. It definitely doesn’t have any clue what Christmas carols are.”

At first, because the network is using random rules to write, it spits out rubbish, like: a a a na snn aca naa i s a i aa a a n a uugna nn na i a uaa a a i a a a a a sna aagt o aa

But after checking the rules against real carols in its database it makes tweaks to improve them, learning how to cap up letters, add in line breaks and figuring out real words.

“Neural networks tend to pick up tone and vocabulary pretty easily, but struggle with making sense,” Shane said.

The resulting carols have a certain rhythm to them but make no real sense – much like poems that literary types might describe as having hidden meaning and depth.

Choice couplets include:

And later:

There are also some more, er, explicit takes that Shane cut out of her blog for reasons of good taste – obviously not something that so troubles El Reg.

Merry Christmas! ®