Aboriginal language Jingulu, spoken by the Jingili people in the Northern Territory, could help solve complex AI problems, Australian researchers led by UNSW Canberra’s professor Hussein Abbass has found.
Its characteristics allow it to be easily translated into AI commands, potentially helping to solve some of the most challenging communication problems between humans and AI systems.
Although other Indigenous languages could offer the same potential, it’s Jingulu’s structure, with its reliance on three light verbs only, that offers so many different applications in AI systems.
It can be used to develop commands to direct movement of single- and multi-robot systems in many different applications, Professor Abbass told Information Age.
The three verbs – come, go and do – make it unique and able to effectively communicate spatial movements.
This simple, efficient syntax could reduce computational cost in AI systems and has been shown to work in situations where communication between humans and a large number of AI agents is required.
“This offers extreme flexibility in computer communication when words could arrive through different routes and the sentence can still be understood without paying extra cost in ordering it,” Professor Abbass said.
The research team recognised this is another example of Aboriginal people’s long history of contributions to the defence of Australia.
“During the Second World War, their languages were used for secret communications. Today we’re discovering that the wealth and richness of the Aboriginal languages and culture could hold the secret in human-AI interaction.”
Solving complex problems with simple AI commands
Professor Abbass works with swarm systems, where groups of robots (or AI agents) work together to solve very complex problems or perform tasks.
His systems draw their inspiration from sheepdogs, where a few sheepdogs can control a large flock of sheep.
“This problem is all about movements in different information and knowledge spaces, including the physical spaces,” Abbass said.
“These movements are represented mathematically as elements that get attracted to each other or repulse from each other. For a long time, I have been looking at how we can design the languages used at the interface between the swarm and humans.”
Professor Abbass has investigated systems that rely on gestures, direct commands and even music, but they all had their challenges.
“They either had a richer language than what we needed or did not map exactly to the mathematics we use for guidance and control,” he said.
This all changed when one day, out of curiosity, he was searching on Google for studies that looked at the syntax of Aboriginal languages.
“I encountered a PhD thesis about Jingulu. I started reading it and then it didn’t take much time before it clicked in my head — this language would be perfect for my artificial intelligence-enabled swarm guidance work.”
The structure and makeup of Indigenous languages, which are both non-configurational and display free-word order, results in the meaning being preserved even when syntactic and word orders change.
Working alongside University of Canberra linguistics expert associate professor Eleni Petraki and defence science and technology group Dr Robert Hunjet, the team created JSwarm, a language inspired by Jingulu.
Finding other applications
Jingulu can change the context of use without changing its syntax which allows the AI to be transferred between different domains with ease.
Professor Abbass praised the language which can translate straight into AI commands and as a human language, humans can master its full understanding, paving the way for widespread adoption.
“It’s a language that is born and used in Australia to support research and innovation that are born and used in Australia,” he said.
This is just the start and the researchers plan to widen the scope of applications.
“I’m looking for funding and collaborators who would be willing to collaborate with us to transform this technology into multi-purpose application in different domains,” Abbass said.
