Te Kete o Karaitiana Taiuru (Blog)

This week Google Maps began rolling out a New Zealand voice that pronounces te reo Māori place names correctly. Place name Whangārei with the soft “f” sound from an New Zealand dialect, Taupō with the long vowel, Karangahape Road spoken as it was named by our tūpuna, not as decades of navigation systems and generations of non Māori descendants being culturally ignorant New Zealand’s Māori culture and the inter-generational impacts of cultural assimilation, have butchered our language for decades and even centuries.

 

Source: https://blog.google/products-and-platforms/products/maps/te-reo-maori/

 

What Google and Te Taura Whiri have built

The new voice is an AI text-to-speech model developed by Google in partnership with Te Taura Whiri i te Reo Māori (The Māori Language Commission established under the Māori Language Act https://www.legislation.govt.nz/act/public/1987/176/en/latest/#DLM124146 ). Their relationship stems from a Memorandum of Understanding signed in 2023, which earlier produced a localised Chromebook with a reo Māori keyboard.

For the Maps voice, Te Taura Whiri linguists worked directly alongside Google engineers over several years. The model was trained on recordings of a New Zealand voice actor with both a Kiwi English accent and strong te reo pronunciation, which is what allows the voice to move from an English instruction into a Māori place name without the jarring shift we have all learned to expect. The pronunciation rules were guided by Te Taura Whiri alongside publicly available New Zealand Geographic Board data.

The launch covers cities and towns, with streets and roads to follow. Te Taura Whiri stated the technology will not be perfect from day one, so has set up a public channel for reporting mispronunciations so the model can be taught the right way to say a word here.

That correction pathway keeps Māori expertise in the loop rather than leaving our language to Google’s internal triage as what occurred with Google Translate Māori, a project lead by Māori Language Tech expert Te Taka Keegan and Google in 2009 that revolutionised reo Māori revitalisation via online translations, eventually leading the translation tool to be substandard to language experts due to no ongoing oversight and investment.

 

Digital Colonisation

The default treatment of te reo Māori by the technology industry continues to be extraction without consent and without governance. Our language already sits inside the training data of the major large language models, scraped from digitised archives, YouTube, illegally digitised books, journals and the open internet, taken without permission from the descendants of the people who kept the reo alive through active Crown suppression, with no kaitiaki, no right of correction, and no benefit flowing back. Te reo Māori is a taonga guaranteed protection under Article 2 of Te Tiriti o Waitangi, affirmed by the Waitangi Tribunal in WAI 11 and again in WAI 262. Unfortunately, the lack of copyright and other Intellectual Property laws both in New Zealand and internationally, has given industry free range to treat our taonga as free raw material that legally no one can individually claim ownership of.

The kaitiakitanga arrangement

Google has stated that Māori data sovereignty was central to the development of this model. Te Taura Whiri will initially act as kaitiaki (custodian) of the Māori lexicon behind the voice. The longer-term plan is to establish a group of Māori custodians, people with expertise in both te reo and information technology to hold guardianship of the data, so that Māori academics, researchers and communities can access it, share in its benefit, and shape where it goes next. Notably, the intellectual property is owned or protected by Te Taura Whiri at this stage.

An arrangement in which the Māori lexicon has a named Māori kaitiaki, a planned transition to broader Māori custodianship, and a commitment to Māori access and benefit is genuinely different and alights fully with tikanga Māori. This may be the reason commercial te reo providers couldn’t make a deal in the past.

This kaitiaki arrangement proves the point I have been making to government agencies and companies for years: Māori data sovereignty is not a barrier to innovation. It is a design requirement, and any serious partner can meet it when it chooses to. Google has chosen to in this example creating a benchmark that now exists against which other big tech and government can be measured.

Three distinctions in this arrangement need to be constantly reviewed, because they will determine whether this becomes a true precedent or a well-intentioned exception.

1. The public statements describe Te Taura Whiri as kaitiaki of the Māori lexicon, the pronunciation data and linguistic rules. The trained model itself, the infrastructure and the commercial product remain Google’s. The reo goes in as taonga under Māori guardianship; the capability it creates comes out as a corporate asset. The test of this partnership is whether benefit-sharing extends to the value the model generates, not only to access to the underlying data. If it does not, we have refined the extraction, not ended it.
2. Te Taura Whiri is a Crown entity that Ngahiwi Apanui-Barr said it himself: Te Taura Whiri does not own the Māori language, Māori people do. The framing of the Commission’s role as interim guardianship reflects that honesty. Noting te reo Māori is not the property of any commercial entity that is either Māori owned, part Māori owned or non-Māori owned and it is not owned by an Iwi or iwi representative.
3. Under Te Tiriti, tino rangatiratanga over taonga sits with hapū, not with a Crown agency, however capable its leadership. The planned custodial body is the mechanism by which this arrangement either honours Article 2 or quietly substitutes Crown stewardship for Māori authority. Its composition, mandate and independence will decide which.

The Māori language is a living taonga that has been gifted by Māori language activists to the country and the world when it was made an official language of New Zealand. For anyone to try to claim the language either as an individual representing all of Māori Peoples or by any entity is mischievous and misrepresentation. Extractivism in reverse.

In relation to my State of the Nation of Māori Data Report 2025 and subsequent 2026 research to date, suggests that this is the first Crown organisation to operationalise Māori Data Governance within the whole of the Crown sector. Te Taura Whiri are also one of the least funded Crown agencies.

Place names

Our place names are not generic as they carry the dialect and the whakapapa of the hapū, marae and whānau whose places and other geographical locations they name, and in some cases, pronunciation is itself a matter on which different communities hold different positions about their own names. A single national pronunciation model, however, expertly guided, makes choices. The feedback mechanism helps, but the deeper question is whose authority resolves a disagreement: the Commission’s, Google’s, or the hapū whose tūpuna named the place.

What this means for the language

Place names are the most public face of the Māori language. They are spoken millions of times a day in cars, on phones, to tourists, to children in the back seat and our English Dictionaries contain thousands of Māori words and names that have become a part of New Zealand English over the past 2 centuries. Every one of them carries whakapapa, tīpuna, events, the relationship between people and land. Correct pronunciation is the doorway to those stories, and until now the most ubiquitous voice in the country walked past the door.

Fifty years of revitalisation from the Māori Language 1972 petition, through Kōhanga Reo, to official language status in 1987 has been about making te reo a normal, everyday language. Normalisation does not happen only in classrooms and on marae. It happens in the mundane infrastructure of daily life. When the navigation app used by nearly every household in the country says Ōtorohanga correctly, it does quiet, repetitive, ambient work that no curriculum can replicate. A child learning the reo will no longer have their pronunciation contradicted by the family car – Tower-poo for Taupō.

There is also a corrective justice dimension that I do not want lost. Technology has been an agent of mispronunciation at massive scale. Every GPS voice that flattened Whakatāne into something unrecognisable participated in the marginalisation of our language. This launch demonstrates that the same technology, governed differently, becomes an agent of revitalisation. The variable was never the capability. It was always the governance. That has been my argument for twenty years, and this week a global company proved it in production.

The precedent

A global technology company that initiated an intellectual property and Māori Data Sovereignty conversation with a Māori entity. It accepted Māori guardianship of Māori language data and committed to a transition toward broader Māori custodianship, and to Māori access and benefit while it shipped a world-class product.

So, the next time a government agency tells me Māori data governance is too difficult to build into an AI procurement, or a company claims Māori data sovereignty is incompatible with innovation, my answer will be short: Google managed it – Why can’t you?

The work now is to see the custodial body established with genuine hapū, marae, Mātanga authority, to push benefit-sharing beyond data access into the value the model creates, and to make this arrangement the floor for Māori data governance in New Zealand, not the ceiling. Our reo waited a long time to be heard correctly after systemic efforts by previous and current governments to silence it in any way they could. It should not have to wait as long to be governed correctly.