The number that doesn't explain anything
You're sitting at a table where three generations eat together. The grandparents argue in the old language, fast and comfortable. The parents follow every word and respond in the national tongue. The children smile at sounds they cannot parse. That scene, repeated across hundreds of communities on every inhabited continent, is why linguists estimate roughly half the world's approximately 7,000 languages could fall silent within a century. But the statistic obscures something crucial: languages with nearly identical speaker counts are doing wildly, almost incomprehensibly different things right now. Some are shrinking. Some are holding. A few are actually growing. The number of speakers, it turns out, is almost the least useful predictor of a language's future.
Welsh has roughly 800,000 speakers. Cornish, once declared extinct, has a few thousand revivalists. Faroese sits around 70,000 speakers on a North Atlantic archipelago. Aromanian, a Latin-derived language spoken across the Balkans, has an estimated 300,000 speakers scattered across five countries. All four are minority languages under a million. Their trajectories since the mid-twentieth century could not be more different. Welsh is taught in schools, carries legal weight in courts, and is the medium of a national television channel. Faroese is the official language of the Faroe Islands, has a full university system, and is spoken by essentially all children born there. Aromanian has no official status anywhere, is excluded from most public education, and its intergenerational transmission is fracturing badly. Same category. Opposite outcomes.
The divergence isn't random. It follows a logic, and that logic is institutional.
What institutions actually do to a language
When linguists talk about institutional support, they don't mean a government putting up bilingual road signs and calling it done. Real institutional embedding means a language becomes the medium through which people conduct the unavoidable transactions of daily life: schooling, legal proceedings, healthcare paperwork, employment applications, broadcasting. The moment a language is confined only to the domestic sphere, it starts accumulating what you might call functional debt, a slow bleed across every domain it cedes to the dominant language, leaving children with no incentive to develop vocabulary, register, or confidence in it.
Consider two children born in the same year into Breton-speaking families in Brittany. One attends a Diwan immersion school, where Breton is the sole language of instruction from age three through secondary school. The other attends a standard French state school where Breton is offered as a forty-minute optional class twice a week. By sixteen, the first child can argue, joke, write essays, and read literature in Breton. The second can greet people and count to twenty. Same heritage, same family, same region. The school did most of the work, and the school's institutional commitment was the only variable that diverged.
That is the mechanism. Institutions don't just teach a language. They manufacture the social contexts in which the language is the natural, unremarkable choice, as ordinary as reaching for a glass of water. Without those contexts, speakers face what sociolinguist Joshua Fishman called a domain collapse: the language retreats until it exists only in intimate settings between aging relatives, and then it doesn't exist at all.
The four levers that actually move the needle
Comparing languages across the under-a-million threshold, four institutional factors appear repeatedly in cases of survival versus decline. None is sufficient alone. All four together constitute something close to a minimum viable framework.
The first is legal recognition, specifically recognition that creates enforceable obligations. Welsh survives partly because the Welsh Language Acts created a right to use Welsh in legal proceedings and required public bodies to produce Welsh-language service plans. That's not symbolic. That's a lawyer arguing a case in Welsh and a hospital required to offer Welsh-language appointments. Symbolic recognition, the kind where a language appears in a constitution's preamble but nowhere else, correlates with almost no improvement in transmission rates.
The second lever is education as a medium of instruction, not merely a timetabled subject. Faroese children don't learn Faroese the way French children learn English. They learn mathematics, biology, and history in Faroese. The language carries cognitive weight across every discipline. Research on immersion programs consistently shows that children taught through a minority language develop stronger bilingualism than children taught the minority language as a subject, because the brain associates the language with complex thought rather than rote recitation.
The third is economic presence. This one makes people uncomfortable, but it is probably the most powerful force operating on young adults making career decisions. If speaking Basque fluently opens more doors in the Basque Country's civil service, media sector, and regional businesses, then Basque becomes a rational investment for a twenty-two-year-old. The Basque regional government has spent decades building this deliberately, creating language proficiency requirements for public sector roles. It is, in effect, manufacturing economic demand for a language the market alone would not have sustained.
The fourth lever is contiguous territory. Languages spoken in a geographically coherent community, where speakers form a local majority in at least some municipalities, have a structural advantage that dispersed diaspora communities rarely overcome. Faroese speakers all live on the same islands. Aromanian speakers are scattered across Greece, Albania, North Macedonia, Bulgaria, and Romania, with no single place where they constitute a significant local presence. Geographic dispersal doesn't doom a language, but it makes every other institutional task dramatically harder, because there is no natural catchment area for a school, no obvious broadcast region for a radio station, no constituency dense enough to elect politicians who speak the language.
What people persistently get wrong
The most common misreading is that community passion is enough. It isn't. There are hundreds of deeply passionate communities watching their languages decline in real time, and the passion is genuine and it is not sufficient. Passion produces festivals, dictionaries, online courses, and beautiful social media accounts. Those things matter at the margins. They do not replace a school system.
The second misreading is conflating language learning with language transmission. A language can have growing numbers of adult learners while its pool of native speakers quietly shrinks. Cornish has more learners now than at any point in the last century. It also has essentially no community of native speakers raising children in the language from birth. Learning and transmission are different biological and social processes, and conflating them produces false optimism in community reports and grant applications.
The third misconception, and probably the most damaging, is that digital presence substitutes for institutional presence. A language having an active Wikipedia, a TikTok subculture, and a keyboard emoji does precisely nothing for the Aromanian grandmother who cannot get a doctor's appointment conducted in her language, whose grandchildren attend schools that treat her language as folklore. I find this substitution genuinely pernicious, because it lets institutions off the hook while communities celebrate metrics that change nothing on the ground. Digital visibility is real and it matters for prestige and learner access. It is not a domain in Fishman's sense. Nobody is obligated to use it.
Why the trajectories keep diverging
You'd expect that, over time, languages with similar speaker counts would trend toward similar outcomes. They don't. The divergence is actually widening, and the reason is compounding. Languages that secured institutional footholds in the twentieth century used those footholds to produce a generation of fully competent speakers, who then staffed the institutions, who then produced the next generation. Welsh-medium schools in the 1970s were small and struggling. By the 1990s, their graduates were teachers, journalists, and civil servants. By now, Welsh-medium education serves more than a quarter of all primary school children in Wales. The institutional investment compounded like interest, quietly and then all at once.
Languages that failed to secure those footholds in the same period entered the opposite cycle. Fewer institutions meant fewer domains meant lower prestige meant fewer parents choosing to transmit the language meant fewer children meant even less justification for institutions. Aromanian's situation today is not simply the result of current neglect. It is the accumulated result of decades of institutional exclusion across multiple national contexts, each year making the next year harder.
And yet the default assumption in policy circles remains stubbornly optimistic about numbers. Have you ever caught yourself thinking that a language with 300,000 speakers is basically fine? That's the trap. The number is almost irrelevant without knowing whether those 300,000 speakers are distributed across schools, courts, broadcasters, and employers, or whether they are 300,000 elderly people whose grandchildren speak something else entirely.
The uncomfortable arithmetic of survival
There is a floor below which institutional intervention probably cannot reverse a language's decline within a single generation. Linguists argue about where that floor sits, but most place it somewhere around the point where children are no longer acquiring the language natively in significant numbers, regardless of how many adults still speak it. Below that floor, you're doing reconstruction, not preservation. Reconstruction is possible, as Hebrew's twentieth-century revival demonstrated in extraordinary circumstances, but it requires a confluence of political will, territorial concentration, and ideological commitment that is genuinely rare.
Above that floor, the evidence is genuinely encouraging for languages with institutional backing. Welsh speaker numbers have grown since comprehensive language policy took hold. Basque transmission rates among children in the Basque Autonomous Community have improved measurably over several decades of regional language planning. Faroese has not merely survived: it has developed a full literary culture, a film industry, and academic publishing in a language with 70,000 speakers on islands the size of a mid-sized city.
Small languages are not fragile by nature. Some of the most resilient language communities on earth speak languages most people have never heard of. Resilience is built, not inherited, and it is built in legislatures and school boards and civil service hiring committees. The family kitchen matters. It is just not enough to do the job alone.