We had heard about some of the work happening with First Nations languages and artificial intelligence around the world and wondered: What is it? Could it offer ways of practicing or preserving our language? Can a machine understand Arrernte? And whose place is it to teach the machines? The future of our language lies in our young people. They will be the ones to make decisions about our language, so we wanted them to understand how this technology works.

These students are our future language leaders. As part of a Certificate II in Applied Languages, the program at Centralian Senior College is all about building pathways, showing young people that their language is something that offers job opportunities as well. The students come together four times each week from different schools around Mparntwe and many speak multiple languages like Eastern and Central Arrernte, Western Arrarnta, Pitjantjatjara and Anmatyerr.

We began exploring AI bossing creative technologist/producer Caddie Brain dressed up as a robot around the classroom. We thought robots were smart, but it turns out you have to tell them exactly what to do. You have to program them. But what with? And how? We learned about data by getting a box of Smarties (a kind of lolly or candy). We classified them by colour and arranged them into data visualisations. We then used Teachable Machine to train our computers to recognise those colours. Then we ate the data.

Our language is data too. We began to experiment with building a dataset from hundreds of recordings of the different Arrernte animal names. We had to record ourselves saying the words over and over and over again to train a model. Bit by bit, the computer began to recognise our words as we spoke to it. For example, if there were too many boys' voices it had trouble understanding when a girl was speaking. We learnt that this is called bias in datasets. We also accidently mispronounced one of the animal names, so we had to start again. (‘Aherte’ and ‘arerte’ are similar but not the same. One is a bilby, the other means ‘crazy’!) We need our language bosses, people like Veronica Perrule Dobson, Kathleen Kemarre Wallace and Joel Perrule Liddle, to check our work and help build the dataset, making sure it is exactly right. It made us think more about what the cultural protocols might be to train a dataset like this. For that reason we have decided not to share our dataset for now. While we want to share our language, we also want to care for it and know what is done with it and our voices. Data sovereignty is important to us. 

We then designed the game, working with graphic designer Graham Wilfred Jnr. He took inspiration from arcade games and brought this to our landscape. We asked Western Arrarnta country singer Warren H Williams (we’re big fans!) to make us some sounds for the game. He recorded them on a keyboard for us. We then built a P5 sketch with our graphics, sound and Teachable Machine model. Yining Shi and the Processing Foundation worked with us every week helping us make changes and to try things out. And the game was made. 

While we had learnt so much about machine learning dreaming, we realised that we were only understanding it one way, through English. There is no Arrernte word for computer for example. So to truly consider these ideas, we needed to look at them through an Arrernte lens, to understand them our way. So one day in class we spent hours trying to translate key concepts into Arrernte. Could ‘computer’ translate as ‘to type’ or is it a kind of ‘a thing’? What algorithms hide in our language? What metaphors are there for data? Is it like sand? Is ‘training’ a machine the same thing as ‘teaching’ it? Is artificial intelligence even smart? Or is it ‘fake’ knowledge?

We took these ideas to Arrernte linguists and emoji bosses Veronica Perrule Dobson, Kathleen Kemarre Wallace, Joel Perrule Liddle who spent many weeks thinking through what these terms should and could be. There are of course, no direct translations for them. 'Akurrknge alharrkentye' the term of computer, translates as ‘lightning brain’ for example. This is old language. This is deep language. Every word deliberate. Every word existing in the closest relationship with the one next to it to build meaning. They are listed below. You can listen to them also. Listen to understand.

They also had another message for us.