Probable Words

Max Neupert Probable Words - Artist's book


Probable Words is an artist's book with poems based on the next-word predictions of the Android keyboard SwiftKey®.

Probable Words
Max Neupert
Self published in 2016
108 pages. 9.7 x 14.7 cm
60 numbered copies


SwiftKey®, like other available keyboard apps, proposes the most likely words to follow the previously typed one. SwiftKey® was a British technology start-up, founded in 2008. It has been acquired by Microsoft for US$ 250 million in 2016. The poems begin with a word that either came to my mind out of the situation, or a friend chose for me. From there, every word is one of the three words suggested by SwiftKey®. Their choice is deliberate, but spontaneous.

The next-word choices provided by the software are based on statistical data. It continuously learns what the user types and incorporates this into the predictions. To remove this growing bias, the keyboard was reset to its defaults periodically. Because the following word is always one of the statistically most probable words to come next, every word-pair in the poems feel familiar and seem to makes sense. However, the whole sentence usually is grammatically incorrect and logically flawed. Auxiliary verbs, negations and prepositions like "of", "with", "to", "from" and so forth are amongst the most likely words to succeed any word. By following the word suggestions the sen- tences thus become long and convoluted.

The poems are a hybrid of stochastic prediction and conscious choice, seasoned with my personal writing style on the phone. It's neither random, nor carefully phrased, but rather crudely mended to- gether from the tight restrictions of the three choices.

There is nothing "smart" or advanced about the word predictions. The algorithm is entirely agnostic towards sentence structure and grammar. It simply proposes word successions from a pool of words. However it provides a window into the possibilities of artificial intelligence and its resulting poetry holds up a mirror of our desire to give sense to gibberish.

On Nov. 20, 2015 I tweeted to @SwiftKey that I want to train SwiftKey® on Shakespeare texts, to make it suggest words that Shakespeare would have written. On April 13, 2016 SwiftKey® announced an app Shakespeak, that does exactly that.

The Twitter account @probablewords was used to collect the poems in English, Korean and German.


Download the PDF versions of Probable Words.


Print by Daegu Print
Layout based on Philipp Reclam jun. Stuttgart, used without permission
Typeset in Scribus, Font: Garamond No. 8
SwiftKey is a registered trade mark copyright by TouchType Ltd 2016, a subsidiary of Microsoft.
This book is in no way affiliated with SwiftKey, TouchType, Microsoft or Reclam.
It does not endorse them or any other commercial input method in any way.
Installing a third party keyboard on your device is a possible attack vector and may breach your privacy.
I recommend the open source F-droid repository and the AnySoftKeyboard.


Page updated: 2016-05-12

Valid XHTML 1.0 Transitional