Vaulting the Data while Counting only 4% of Kiwis

In July 2026 Google Maps began saying Te Reo Māori place names correctly to drivers across Aotearoa. Te Taura Whiri i te Reo Māori, the Māori Language Commission constituted in 1987, built the text-to-speech model with the company across several years of phonetic detail. The recordings came from a voice actor with a Kiwi English accent and strong Reo Māori pronunciation. Against the published record of place names, Commission staff wrote the pronunciation rules themselves. Cities and towns came in the first release, streets and roads in a later batch.

Under the terms published so far, Te Taura Whiri exercises Kaitiakitanga over the Reo Māori data behind the voice. Google's licence is for this text-to-speech model only. In the longer term, according to the Commission, a consortium of interested groups will assume the Kaitiakitanga. Māori researchers and communities would then draw on the data and share in the benefit. With the Commission comes the Intellectual Property protection, at least at this stage, as mispronunciations will be collected via public consultation for correction into the model.

Te Taka Keegan, a Reo technologist, led the 2009 project adding Māori to Google Translate. For learners it was the first online lookup in the language. Experts later judged the tool substandard, the model left for years without oversight or investment. By 2023 a memorandum between the Commission and Google had already produced a Chromebook with a Reo Māori keyboard. We still argue that generative systems underperform with low-resource languages for lack of good-quality data, essentially.

Since 2023 Te Kāhui Raraunga has built the second half of the programme. Its Māori data governance model, published that year, addressed a government readership. In May 2026 the organisation launched Te Pā Tūwatawata, a decentralised storage network open to marae and iwi, owned and operated by Māori engineers. Principal advisor Erena Mikaere put the case bluntly: Without command of its digital infrastructure, a nation has a sovereignty stopping at the server port.

In the 2023 census, 213,849 people reported a conversation-level command of Reo Māori, against 185,955 five years earlier. That is 27,894 more speakers, a rise of 15% and a conversational base of 4.3% across Aotearoa. Gisborne returns the highest regional proportion at 16.9%, Northland the next at 10.1%. Within the Māori population the proportion has barely moved, 18.4% in 2013 against 18.6% in 2023. Te Kāhui Raraunga counts 49% of iwi speakers under the age of 30.

Because the internal proportion has stayed flat while the absolute count climbs, most new speakers are being exposed to Te Reo Māori outside the Māori population. Confining a licence to Māori academic and community use protects the corpus and leaves a mainstream school without a text. Kaitiakitanga will not by itself lift the 4.3%. At that scale, growth needs the 95.7% with no conversational Reo. The 49% of iwi speakers under 30 will find them in the schools more than in the repositories.

An earlier commission put comparable wiring through a different instrument. A digital-services provider needed its privacy notices and personal-data consent statements in Samoan as well as English for a grant. Our teams pulled the grant back to the purpose stated by the operator, then rewrote the rights summary at plain reading level. A named national service line stayed in its English form at the client's instruction.

Taonga is promptized into the model under Kaitiakitanga. Out of it comes the trained model and its precious weights, corporate assets under Google's control. Benefit sharing measured against data access refines the extraction and does not end it. According to Ngahiwi Apanui-Barr, the Commission does not own the Māori language. Ownership belongs to Māori people. A Crown entity administers it and, under Te Tiriti, tino rangatiratanga over taonga belongs to hapū.

For procurement teams weighing a speech product for Reo Māori, a vendor claim proves nothing. Huri Translations posted an LLM comparison table for Māori against a mainstream engine, with our own graded evidence. In practice, vendors describing a Pacific voice as native-quality are describing a training set, really.

Carriers briefing cabin crew on New Zealand destinations want the pronunciation data. So do publishers building classroom audio for learners in Sydney or Brisbane. Our Auckland hub fulfills these requests from operators with no Māori staff and no path to a licence. Under the current terms an overseas publisher has no lawful source for the lexicon. The remaining option is a scrape of the open internet, the practice targeted by the sovereignty programme.

In contrast, in a 2017 report the French Polynesian culture ministry counted more than 12,000 ʻOri Tahiti dancers in the United States and more than 10,000 across Latin America. Japan had 25,000 against a projection of 500,000 by 2027. Against fewer than 50,000 dancers in Hawaiʻi itself, Hawaiʻi Tourism Japan puts the Japanese Hula figure at roughly 2 million. Yep, TWO MILLION Hula dancers in Japan. That spread happened without a licence and without a vault. Out of the islands went songs, chants and repertoires with their kumu and arataʻi.

Throughout the diaspora the island practitioners remain the reference. Tumata Robinson and Manouche Lehartel founded the ʻOri Tahiti Nui competition in 2012, judged by Tahitian professionals. From the Tōkyō and Los Angeles Heiva the winners qualify for the World Championship in Papeʻete. In French Polynesia, an old 2012 survey found 73.5% of residents competent in a Polynesian language. Hawaiian shows roughly 300 first-language speakers against 22,000 second-language speakers.

To serve revitalisation, a licence needs published vectors: Purpose limitation, but also Education and public broadcasting. Those are permitted uses. To the naming Hapū goes the attribution. FIdelity to spellings is in the contract, since macrons must survive the delivered files and the vendor's own tooling as well. But versioning matters as much, because a lexicon corrected in 2027 and a product shipped in 2026 will diverge silently.

In July 2026 Te Taura Whiri stated the transition as a long-term goal. Still outstanding is the licence schedule that goes with it. 50 years of revitalisation since the 1972 petition put