Safeguarding Vagahau Niue: Exercising Linguistic Sovereignty in the Age of AI

The issue is not merely whether artificial intelligence produces inaccurate language. It is who holds the authority and expertise to verify, correct and shape the digital representation of Vagahau Niue, the Niuean language.

A phrasebook may appear modest. Yet when a book claims to teach a language and contains material that fluent Niuean speakers describe as inaccurate or drawn from other languages, it raises a fundamental question of linguistic and informational authority. Niuean institutions are now confronting this concern.

What is known, and what remains uncertain

According to Pacific Media Network, the Niue Language Commission and Niue’s Crown Law Office have examined Niuean to English Phrasebook: Everyday Common Words and Phrases (Phoenix Sun Publishing), offered through online booksellers including Amazon and AbeBooks. Niuean educators who reviewed pages shared with them said the material appeared to misrepresent Vagahau Niue and to combine Tongan with other Pacific languages.

Concerns were raised that the book might have been generated using artificial intelligence. But no publicly available technical assessment has established how it was produced or whether an AI model or another automated process was involved. If generative AI was used, the case illustrates how unreliable output can acquire the appearance of a credible educational resource without adequate linguistic verification. Even if no AI was involved, the underlying information failure remains.

The central problem is not simply that a system may have “hallucinated”. It is how doubtful content can be generated or collected, selected, packaged, marketed and distributed as knowledge.

Information disruption

In my research on legal information practices, I use the term information disruption for situations in which necessary information cannot be obtained, or an important source supplies inaccurate, irrelevant or unusable information that obstructs proper performance. The concept applies here.

A phrasebook containing materially inaccurate or unverified content can become a disruptive information source. Learners may rely on it; teachers may have to correct it; search engines may index it; and later publications or AI systems may reproduce it without evidence that fluent Niuean speakers validated it.

Three connected failures may arise. Content fails when material presented as Vagahau Niue is linguistically inaccurate. Use fails when uncertain material is relied on for learning without sufficient verification. Governance fails when publishing and distribution systems give such material reach and an appearance of authority without adequate review, transparency or correction.

The failure cannot be attributed to an algorithm alone. A machine may produce an incorrect translation, but human and institutional decisions can turn such output into a commercial language guide. The relevant unit of analysis is the whole information process, from generation and selection to review, presentation, distribution and reuse. Each stage should be a checkpoint, not merely a conduit.

Why digital scale matters

This risk is especially significant for Vagahau Niue. Tāoga Niue has itself examined the language’s long-term vitality, drawing on Niuean census data and stressing that responsibility for sustaining the language rests with Niuean people themselves. More recently, Niuean language advocates have pointed to strong interest among younger generations in reconnecting with Vagahau Niue alongside shortages of teachers and other language resources.

The quantity of structured and reliably labelled digital material available in Vagahau Niue appears limited compared with English and other widely represented languages. Where structured and reliably validated material is comparatively scarce, an automated system may blur distinctions among related Polynesian languages, misidentify the language or generate expressions that sound plausible to a non-speaker but are wrong in Vagahau Niue.

The consequences can extend beyond the digital environment itself. A book sold online may also enter homes, classrooms or library collections in print. Inaccurate material can therefore become part of the limited body of resources through which learners encounter the language and from which later publications or automated systems may draw. Vagahau Niue is a living language carried through speakers, families, education, songs, stories and cultural knowledge. Its digital and physical representation should remain connected to the people, knowledge and institutions through which the language is sustained.

The effects are also uneven. A global marketplace may see one disputed book as a minor listing. For Niuean educators and institutions, it concerns the integrity of a language with statutory, cultural and public importance. Distribution is rapid and inexpensive, while identifying and correcting linguistic errors requires local knowledge, attention and resources.

Through the Language Commission and Crown Law Office, and within the broader work of bodies concerned with Niuean language and heritage, Niue is exercising institutional responsibility for protecting and supporting the accurate representation of Vagahau Niue. It is also articulating the standards by which materials presented as educational resources should be assessed.

Accuracy is not enough

Accuracy must be judged against purpose. Guidance from the Spanish Data Protection Agency, issued in the different context of personal-data processing, offers a useful principle. The quality of data used in AI depends not only on whether individual items appear correct, but also on provenance, representativeness, context, bias, outputs and suitability for the intended use.

A phrasebook is not fit for teaching merely because some entries appear fluent. What matters is whether its sources, review process and final content are reliable enough for language education. Material suitable for experimentation may be unsuitable for publication with the apparent authority of a teaching resource.

Linguistic sovereignty and Niuean agency

This leads to linguistic sovereignty: the capacity of a community and its legitimate institutions to participate meaningfully in how its language is documented, classified and taught, and to challenge materially inaccurate representations. Niuean institutions, speakers, educators and members of the diaspora have an active role in assessing materials presented as teaching or authoritatively representing Vagahau Niue. In this case, that is precisely what happened: an individual reader identified serious concerns, prompting closer linguistic scrutiny.

Reliable language resources are part of the information infrastructure through which people in Niue and across the diaspora learn, communicate and maintain their relationship with Vagahau Niue. Responsible use of AI need not threaten that. Tools developed according to Niuean priorities and informed by competent speakers may support teaching, transcription and intergenerational transmission. The point is not to exclude technology, but to keep the digital representation of the language connected to those who know it.

That approach is consistent with UNESCO’s Global Action Plan for the International Decade of Indigenous Languages (IDIL 2022-2032), which addresses artificial intelligence and language technologies and calls for Indigenous Peoples themselves to be involved in standardisation and content development.

What responsible publication requires

Educational material produced substantially through AI should disclose that fact and state whether it underwent meaningful linguistic review. Online marketplaces should provide an effective process through which recognised Niuean institutions and qualified language experts can report serious inaccuracies and receive a reasoned response.

Where material is presented as authoritative in Vagahau Niue, publishers and marketplaces should give substantial weight to the standards and assessments of Niuean institutions, speakers and educators. That is not an unrestricted veto over publication. It is recognition that linguistic authority requires more than fluent-looking text and a commercial listing.

There is also a practical lesson for Israel. AI performance should be assessed by language, community and purpose, not only through aggregate multilingual results. A system that performs well across widely represented languages may still fail in context-sensitive uses of Hebrew or Arabic, while the risk may be greater for less digitally represented languages.

Niue should not have to choose between limited digital presence and external misrepresentation. Its population size does not diminish the authority or expertise of its institutions, educators and speakers to assess resources presented as teaching or representing Vagahau Niue.

The question is not whether AI may speak Vagahau Niue. It is whether those who design, publish and distribute linguistic content are prepared to recognise Niuean authority and expertise in setting the standards by which such material should be judged.