Shakespeare in Swahili: Language, Power and the Layer Africa Can Own

A president who translated Shakespeare understood something most African AI strategies still miss: whichever language carries a nation’s highest thinking becomes the language of its power. 

In the age of AI, the continent’s two thousand languages are either its widest moat or its quietest surrender.

DAR ES SALAAM: THE PRESIDENT WHO TRANSLATED SHAKESPEARE

Julius Nyerere was a schoolteacher before he was a president. His people called him Mwalimu, the teacher, long after he had stopped teaching anyone but a nation. And at some point in the crowded years after Tanganyika’s independence in 1961, while running a new state with almost no trained administrators, he found the time to do something no head of government on earth would have thought urgent.

He translated Shakespeare into Swahili. First Julius Caesar, published as Juliasi Kaizari. Later The Merchant of Venice, which he renamed Mabepari wa Venisi: the capitalists of Venice.

It was not a hobby. It was a demonstration. Colonial education had taught a generation that serious thought lived in English and that African languages were for markets and kitchens. 

Nyerere set out to prove, line by line, that Swahili could carry the most celebrated political drama in the English language: conspiracy, rhetoric, the assassination of a ruler, the fickleness of a crowd. If Swahili could hold Shakespeare, it could hold a parliament, a curriculum, a nation.

Tanzania made Swahili the language of its schools, its government and its national identity. Today the United Nations counts more than 200 million speakers of the language and marks a World Kiswahili Language Day every 7 July.

Nyerere understood something that most African AI strategies, sixty years later, still miss. Whichever language carries a nation’s highest thinking becomes the language of its power. The machines are now deciding which languages those will be. 

Africa owns a third of the world’s languages and almost none of the digital record of them. That gap is either a moat or a grave.

IN BRIEF

■  Africa speaks about 2,000 of the world’s 7,000 languages, but African languages make up under 0.1% of online content, the raw material of AI.

■  Language is the one layer of the AI stack that cannot be bought abroad, because it is inseparable from the people who speak it. That makes it Africa’s best moat.

■  Nigeria’s N-ATLAS and Lelapa AI’s InkubaLM show the work has begun. The decisive question is who owns the models, the data and the dictionary.

THE PROPOSITION: African languages and AI: the moat that could become a cage

Here is the paradox at the heart of AI in Africa. The continent’s linguistic diversity is routinely described as an obstacle. It is in fact the only layer of the AI economy that Africa can own outright.

Last week this column described the four grounds on which the AI leap must land: power, compute, language and data, and the rails of payment and distribution. Power takes decades and capital Africa often lacks. Compute is bought from a handful of foreign suppliers. But language cannot be manufactured in California or Shenzhen. It lives in mouths, markets, mosques, churches, courts and family arguments. It is the one input to AI that must be acquired from Africans, on African ground.

That is why the world’s largest model makers are now quietly hungry for African speech. A model that can bargain in Hausa, counsel in Yoruba, triage a patient in Amharic and explain a loan in Swahili will own the next billion customers. Africa’s languages are not a barrier to that market. They are its gate.

Language is the only layer of AI that cannot be imported. Whoever owns Africa’s languages in the machine will own Africa’s customers in the market. – David Adeoye Abodunrin

THE INVISIBLE ROT

The prestige reflex: why Africa’s elites are training the machine to ignore them

The rot here is behavioural, and it begins at the top. In most African boardrooms, ministries and universities, serious business is conducted in English, French or Portuguese. Mother tongues are for the village, the joke, the prayer. I call this the prestige reflex: the instinct to switch into a colonial language whenever the stakes rise.

For a century the reflex cost Africa dignity. In the age of AI it costs something more measurable. AI systems learn from the text the world produces. When a continent’s elites write their contracts, policies, research, reports and customer service in borrowed languages, they generate almost no digital record of their own. The machines then learn, correctly, that African languages are not where important things happen. And they serve African citizens accordingly: fluently in English, haltingly or not at all in the languages those citizens think in.

The cost lands on the people with least power: the farmer who cannot read an English advisory, the patient who cannot describe a symptom to an English-only chatbot, the trader shut out of an automated credit service because she banks in Pidgin. A digital economy that speaks only the language of the elite will deepen every inequality it touches.

 

THE LENS

The Mother-Tongue Moat

The Mother-Tongue Moat: in a world of borrowed models, a people’s languages are the one asset that must be bought from them. A nation that builds, owns and governs AI in its own languages holds a moat no foreign model can cross without paying the toll.

A moat is not a wall. It does not keep foreign models out; it decides the terms on which they come in. If African institutions own curated speech and text corpora, local-language models and the standards that govern them, then every foreign provider that wants to serve the African market must license, partner or pay. If they do not, African languages become free raw material, scraped, translated and sold back.

Most African languages sit on the first rung. Strategy is the decision to climb.

Most African languages today sit on the first rung of the ladder: translated by others. A few are reaching the second, appearing in global datasets.

The goal is the third and fourth rungs: native models trained by African institutions on data they govern, and eventually exported models the world licenses to reach African markets.

WHO WRITES THE DICTIONARY

Power, behaviour and the governance of meaning

The work has begun. The Masakhane research community has spent years building natural-language tools for African languages. Lelapa AI, in South Africa, released InkubaLM in 2024, a small language model for several African languages. 

In September 2025, at the United Nations General Assembly, Nigeria launched N-ATLAS, an open-source model for Yoruba, Hausa, Igbo and Nigerian-accented English, developed with the Lagos firm Awarri.

The models are arriving. The decision that matters is who owns them, and on whose terms they speak.

Power – These efforts matter less for what they can do today than for who controls them tomorrow. The entity that trains the dominant Yoruba or Swahili model will decide, in practice, how those languages are spelled, which dialect counts as standard and which words are flagged as offensive. That is the power to write the dictionary, and a dictionary is a constitution of meaning.

Behaviour – Adoption will follow trust, not quality alone. Speakers who have been taught that their language is second-class will first test any African-language AI with suspicion. A model that gets idioms, proverbs and courtesy wrong will be mocked and abandoned. Getting respect right is as important as getting accuracy right.

Algorithmic governance – Governments hold the largest store of high-quality local-language text there is: court records, radio archives, school examinations, parliamentary debates, agricultural extension scripts.

Released under the right licences, these become the training data of a national language industry. Locked away, or given to the first foreign company that asks, they become someone else’s moat.

THE WEDNESDAY BRIEF

Strategic answers at three altitudes

THE INSTRUMENT  |  TWO MINUTES | FIVE HONEST ANSWERS

The Mother-Tongue Index

What it measures. How far your own habits treat African languages as carriers of serious work, or as languages for home and ceremony only.

Who it is for. African executives, officials, founders and educators. Score how you actually work, not what you believe in principle.

Email your total, country and role (address at the end). Scores will inform an African benchmark published through The Wednesday Circle.

THE HORIZON

Africa in 2031: spoken to, or speaking