AI being used to analyse thousands of Māori stone artefacts languishing in storerooms around NZ

Wednesday, 22 September 2021

The Southern Pacific Archaeological Research (SPAR) unit form The University of Otago has uncovered a Māori Village in Gisborne that is believed to have been from the 14th century. (report from 2018)

Computational modelling and machine learning are being combined at the University of Auckland to develop tools that will be able to identify and interpret thousands of Māori stone artefacts in storerooms around the country.

“This could enhance our capacity to manage and interpret the heritage record that presently languishes in pieces in museums, labs, and storage facilities,” archaeologist Dr Rebecca Phillipps said.

From that, a more accurate picture of how Māori ancestors were living in the landscape could be created.

The project, which also involves researchers from the University of Otago and Otago Museum will help overcome a critical shortage of resources in archaeology.

University of Auckland archaeologist Dr Rebecca Phillipps with a stone artefact discovered on Ahuahu, Great Mercury Island.

* Archaeologists have a field day on Invercargill block site

* The bay where Māori and Pākehā cultural traditions overlapped

* Māori social networks revealed by obsidian

* Research suggests taro important crop for early Māori diet

An excavation site on Ahuahu, Great Mercury Island

The tools it develops will enable automatic identification, measurement, processing, and interpretation of large numbers of artefacts. The resulting digital repositories of Māori cultural heritage will be widely accessible to iwi and archaeologists around the world.

“We have a huge number of stone artefacts excavated from the past, but we haven’t the resources and expertise to do anything about it – it requires a huge amount of time to go through so many artefacts,” Phillipps said. “And there are only a handful of people in New Zealand who have expertise in this field.”

Trustee of Te Ruapekapeka Trust, Pita Tipene, explains in te reo Māori and English why Ruapekapeka pā site is important.

The project will develop a database that firstly distinguishes the human-modified artefacts from those that have been created by natural environmental forces. It will also involve “teaching” a machine to accurately recognise one from the other and identify the shape and size and characteristics of each object.

Among those involved is Professor Poul Nielsen, who is based at the University of Auckland’s Auckland Bioengineering Institute, where his research normally focuses on mathematical modelling of complex biological systems.

The project was an opportunity to take the institute’s research outside bioengineering and apply it to cultural heritage, he said.

The work also draws on the machine-learning research of Professor Mark Gahegan and his team in the School of Computer Science.

“Using the latest deep learning methods, and high-resolution imagery of the artefacts, we can learn to recognise subtle differences and similarities among artefacts automatically,” Gahegan said.

Multicamera stereoscopic systems developed by Nielsen and his team will be used to scan artefacts, characterise their geometry, and digitally reconstruct the objects from which they came.

Cindy Lilburn from Palmerston North museum Te Manawa with part of an historic adze head found at a gravel yard near the Manawatu River in late 2020. (File Pic)

Artefacts could be categorised in a way that they could be found when wanted for particular areas of research, Gahegan said.

Phillipps said the artefacts in storerooms around the country probably covered all periods following the arrival of Polynesians in New Zealand, although it was generally thought the larger assemblages of artefacts that had been found were from the earlier periods.

The project was focusing on learning about the materials used, how the artefacts were made and what people were using them for, she said.

Those uses covered a wide range of activities, such as cutting up meat, carving wood, carving bone for fish hooks, and cutting down trees.

Contracting archaeologist Andrew Hoffman with a stone adze found at an archaeology dig on a housing development at Whitianga. (File pic)

There were sharp slivers used for say cutting up seals, to tiny artefacts for working with flax, and also many adzes. Probably most common was the debris created during the process of making a tool, Phillipps said.

“It’s interesting to learn how people were making those tools, where they were making them, and where they were sourcing the stone from.”

In the Pacific as a whole there were some fascinating patterns of movement of stone. People were using it, seeking it out and trading it.

That didn’t change when people reached New Zealand. The difference was that this country’s amazing geology provided rich and varied stone sources, Phillipps said.

“It says something about how people were interacting and using the landscape.” There were plenty of examples of experimentation in tool making, such as with adzes. “You (also) see people starting to make things that were previously made from bone or shells, and experimenting and making them in stone.”

Pounamu was not found elsewhere in the Pacific, and its use was probably not widespread in New Zealand until sometime after Māori arrived, and it was not used for everyday utilitarian implements.

“What most people are using every day are obsidian, or chert … a very common middle range kind of rock, very common in New Zealand, quite closely related to quartz, just a bit more robust,” Phillipps said.

While there were only 27 known sources of obsidian in New Zealand, mostly from Taupō through to the Bay of Plenty, artefacts made of the stone were found as far away as the South Island.

“That means either people from the South Island were bringing the material back, or they had trading networks or relationships with other people able to source that material.”

Part of the aim of the project was to create a product that could be used internationally by people involved in archaeology. Colleagues in Australia, where collections were many times larger than those in this country, were interested, Phillipps said.

While some of the artefacts now in collections had been fully studied, many had not been, and plenty more were coming in from such places as construction sites and university projects.

A small pilot using photographs – 2D images – had already been run, and the AI had done well at differentiating between various types of artefacts. Phillipps was confident of getting good results when the AI was provided with 3D images.

Up until now AI and machine learning had been used in archaeology more on a landscape scale, such as for studying satellite images, she said.

And while 3D imaging was common in archaeology, the project was using next level technology. “That level of refinement and detail … you can get from the images is far superior to what we have had.”

The Ministry of Business, Innovation and Employment’s Endeavour – Smart Ideas programme has provided $1 million for the work.