Three recent news stories seem to be speaking about different matters. A study on language models in Southeast Asia. An Indigenous political succession in the Brazilian Amazon. A new package of UNESCO grants for intangible cultural heritage.

Yet together they point to a shared tension. When a living culture enters an external system - an artificial-intelligence benchmark, a national law, an extractive market or an international safeguarding file - something is translated, something is protected and something is left outside. The anthropological question is not only what is preserved or what is measured. It is who has the authority to say what it means.

When AI fails, it does not always fail at language

On 2 June 2026, arXiv published SEA-NLI: Natural Language Inference as a Lens into Southeast Asian Cultural Understanding, a benchmark designed to evaluate whether language models understand culturally situated inferences from Southeast Asia. It is not a simple translation exercise. The dataset works with cultural concepts from eight countries - Cambodia, Myanmar, Malaysia, Thailand, Singapore, the Philippines, Indonesia and Vietnam - and with examples in English and in regional languages.

The underlying idea matters. Many natural-language benchmarks are built from Western examples, then translated into other languages, and end up measuring, above all, whether a model recognises already familiar textual patterns. SEA-NLI tries to do something else. It sets out pairs of premises and hypotheses tied to specific cultural contexts, then asks whether the model can infer, contradict or leave open a relation of meaning.

According to the article, the models evaluated perform poorly overall, especially in categories that require denser cultural knowledge, such as languages, science and technology. The problem is not reducible to a lack of linguistic data. The failure appears when the system needs to understand social worlds, not only strings of words.

The dataset card on Hugging Face adds a key point for an anthropological reading. SEA-NLI is intended to evaluate cultural and multilingual reasoning, but it should not be used to establish essentialist rankings between cultures or to generalise its results to every dialect, register or community in Southeast Asia. The documentation itself warns of relevant limits: partial geographical coverage, synthetic data generated with support from language models and cultural concepts taken from a temporal snapshot, mainly Wikipedia. See the SEA-NLI dataset on Hugging Face.

That is where the value of the news lies. SEA-NLI does not solve the problem of cultural understanding in AI. It makes it visible. It shows that a model can handle grammar and, at the same time, fail to understand the social density that makes a sentence reasonable, ironic, offensive, ritual, improbable or plainly false in a particular context.

For AIthropology Lab, the reading is clear. AI does not fail only when it translates words. It fails when it does not understand worlds.

Inheriting a struggle is not inheriting a post

On 28 May 2026, Reuters published a profile of Megaron Txucarramãe, a 75-year-old Kayapó leader preparing to carry forward the political legacy of his uncle and mentor, Raoni Metuktire, aged 94. The scene takes place in the village of Pykany, in the Menkragnoti Indigenous Territory, in Pará. Reuters places the handover in a context of pressure on the Amazon, illegal mining, deforestation, climate change and restrictive laws against Indigenous territorial rights.

The anthropological interest of the case lies not only in Raoni’s charismatic figure, nor in Megaron’s biographical continuity. It lies in how a form of Indigenous political authority is transmitted. What is inherited is not merely public representation. It is a territorial responsibility, a memory of struggles, a network of political kinship, a relationship with Kayapó youth and an obligation to defend ways of life threatened by state, corporate and criminal actors.

The Instituto Socioambiental describes the Mẽbêngôkre/Kayapó as communities located in scattered villages along the upper courses of the Iriri, Bacajá and Fresco rivers, as well as tributaries of the Xingu, in a territory in central Brazil almost the size of Austria and covered mostly by equatorial rainforest. The profile also stresses the richness of their cosmology, ritual life and social organisation, as well as the intensity and ambivalence of their relations with non-Indigenous society and with international environmentalism.

This dimension avoids a common simplification: presenting Indigenous peoples only as ecological guardians. The defence of the forest matters, but not because the territory is a nature reserve without history. It matters because the territory is also language, kinship, food, memory, childhood, names, ritual, care and authority. The rainforest is not an external resource that can be managed from outside. It is a form of world.

One day after the profile of Megaron, Reuters published another piece on the laundering of illegal gold in the Amazon through mining permits with no real activity. Reuters put the volume associated with those phantom permits at 26.8 metric tonnes of gold, with an estimated value of 3.88 billion dollars between 2018 and March 2026. The primary report by Greenpeace Brasil, Gold Laundering in the Amazon: Anatomy of a Fraud, works with 187 mining titles, identifies 98 garimpo permits with irregularities and counts 25.3 tonnes of gold, valued at R$18.4 billion. Beyond the difference between the two figures, the anthropological point is the same: an administrative architecture can turn territorial harm into an apparently traceable commodity.

The Kayapó political succession, then, is not taking place in the abstract. It is taking place before a system that turns territory into concession, river into externality, gold into administrative traceability and Indigenous culture into a legal obstacle. Against that, Megaron appears not only as Raoni’s “heir”. He appears as part of a historical continuity that insists that Indigenous life cannot be reduced to biodiversity useful to the planet.

For AIthropology Lab, the angle is not “nature against development”. It is more precise and more uncomfortable: who can define what counts as territory, what counts as harm and which forms of life are left out when everything is translated into permits, markets and laws.

Living heritage does not fit entirely inside a form

The third theme seems, at first glance, more institutional. On 9 June 2026, UNESCO announced six new projects supported by the Intangible Cultural Heritage Fund, for just over 320,000 dollars in total. The decision includes three preparatory grants and three international assistance projects in Iraq, Morocco, Samoa, Eswatini, Syria and Tanzania.

Among the examples highlighted by UNESCO are the safeguarding of the Iraqi santur, with 99,840 dollars; the knowledge and skills associated with Moroccan khettaras, with 10,000 dollars; Samoa’s preparatory assistance for the nomination of Tatau Samoa, with 8,049 dollars; Eswatini’s preparatory assistance for a nomination on child-rearing according to emaSwati traditions, with 10,000 dollars; the safeguarding and transmission of traditional Syrian blown glass, with 99,700 dollars; and Tanzania’s project to strengthen the safeguarding of intangible cultural heritage, with 100,000 dollars. The state pages for Morocco, Samoa, Eswatini and Tanzania make it possible to follow part of that cooperation map.

Taken as an administrative note, the news is dry. Read from anthropology, it opens a powerful question. What happens when a living practice needs to become an inventory, a form, a report, an assistance request and a safeguarding programme in order to be protected?

UNESCO’s 2003 Convention defines intangible cultural heritage as practices, representations, expressions, knowledge and skills that communities, groups and, in some cases, specific individuals recognise as part of their cultural heritage. The text of the Convention itself insists on the active participation of those who create, maintain and transmit that heritage. UNESCO develops that idea on its page about the involvement of communities, groups and individuals.

The problem is not that UNESCO intervenes. In many contexts, these mechanisms provide funding, visibility, legitimacy and tools against forgetting, war, displacement, predatory tourism or pressure from states. The problem appears when a living culture is translated too quickly into the language of cultural administration. A practice changes when it is documented. It changes when a “representative” version is selected. It changes when those who speak for the community are named. It changes when it enters circuits of funding, tourism, diplomacy or national prestige.

The anthropology of heritage has long pointed to this tension. Safeguarding is not freezing. Documenting is not possessing. Recognition is not the same as returning authority. And naming a practice can protect it, but it can also fix it, simplify it or turn it into a brand.

Read in this way, heritage is not an institutional appendix, but a third scene in the same cultural operation. The three themes speak of translation. SEA-NLI translates cultural worlds into the language of algorithmic evaluation. Megaron and Raoni translate an Indigenous territorial defence into the language of Brazilian and global politics. UNESCO translates living practices into the language of heritage and international cooperation.

In every case, the common thread is the same: who translates, from where, with what authority and what is left outside the system that promises to understand, protect or represent.

Sources Consulted