On March 1, 2010, Hiroshi Mikitani stood before his employees at Rakuten’s Tokyo headquarters and told them the company was switching to English. Not for international communications. Not for correspondence with overseas partners. For everything. Internal meetings. Lunchroom signs. Emails to the colleague sitting three desks away. The cafeteria menus. All of it.
The target was a TOEIC score of 650. The deadline was two years. Fall short, face demotion.
Sit with what that actually asked of people. These were professionals conducting ordinary working lives in their own country, in their own city, in a company built by and for a Japanese market — and no Anglo-American ownership structure was pressing the mandate down from above. Mikitani chose this. The average TOEIC score among Rakuten staff in October 2010 was 526. The gap between where employees were and where they needed to be was not a technicality. It was a full reorganisation of how they thought, communicated, and conducted the ordinary business of working life. By 2015, the average score had risen to 802.
The question the Rakuten announcement poses is not “why would a Japanese company want to do business in English?” The answer to that one is obvious enough. The question is: what accumulated economic force produces the calculation that restructuring the internal cognitive life of thousands of professionals is less costly than operating outside the English network? Where did that force come from, and why — after the empire that built it dissolved, after the century that carried it forward is now history — does the arithmetic still produce the same answer in Tokyo in 2010 that it produced in Calcutta in 1840?
The merit trap
The most persistent explanation for English’s global dominance is also the least defensible: that English won because it deserved to. The argument takes various forms — that English is unusually precise, unusually expressive, unusually hospitable to new vocabulary, that its stripped-down grammar makes it accessible. None of this survives contact with the evidence.
Consider Esperanto. Ludwig Lazarus Zamenhof designed it in Warsaw in 1887, and by every criterion he identified — regularity, learnability, phonetic consistency, grammatical economy — Esperanto outperforms English. It has no irregular verbs. Its spelling is perfectly phonetic. Its grammar can be learned in an afternoon. After more than 135 years of active advocacy, its speaker count sits somewhere between 100,000 and two million, with the most careful academic estimates placing the true figure of fluent speakers closer to the lower end of that range. Zamenhof thought rational actors would adopt the most efficient communication technology. He was wrong about what rationality produces in language choice.
The Esperanto problem Zamenhof's project did not fail because of insufficient advocacy or imperfect design. It failed because he misunderstood the economics of coordination. A language has value in proportion to the network it connects you to. In 1887, English's network included British imperial commerce, Indian civil service examinations, and American industrial contracts. Esperanto's network was, and remained, only other Esperanto speakers. You cannot build a coordination device by optimising the device. You build one by being where the coordination already happens.
The Latin precedent makes the same point from a different angle. Latin dominated European intellectual life for roughly fifteen centuries. Not because it was clear — it wasn’t, particularly — but because the institutional infrastructure that mattered was organised in Latin. The Church ran in Latin. The universities ran in Latin. When those institutions shifted — the Protestant Reformation democratising scripture into vernacular languages, the printing press making national-language texts economically viable, nascent national scientific academies publishing in the languages of their funders — the language shifted with them. Institutions moved first. Language followed.
The German scientific case is the sharpest historical demonstration. Before the First World War, German dominated natural science. Physics, chemistry, pharmacology: the journals that mattered published in German, and a scientist who couldn’t read German couldn’t follow the field. By the early 1920s, that was over. German and Austro-Hungarian researchers were systematically excluded from international conferences in the postwar settlement — a punishment for the war, and specifically for the deployment of chemical weapons. English-language science didn’t win because it was better science. It won because the institutional mechanisms distributing and validating science shifted to English-speaking countries at a moment when German-speaking countries were locked out. Scott Montgomery’s history of scientific language, Does Science Need a Global Language? (University of Chicago Press, 2013), traces this transition in granular detail.
The mechanism in all three cases is the same. Languages win when the cost of not speaking them rises above the cost of learning them. This has nothing to do with the language’s internal properties.
The East India Company’s language policy
Watch the mechanism operate for the first time in specific contracts and specific resource allocations.
The East India Company’s operational logic, by the early nineteenth century, was not cultural. It was commercial. The Company had reorganised large portions of Indian economic life around English-language contracts enforced by English-language courts. To participate in this commerce — as merchant, clerk, lawyer, intermediary — you needed English. Not because the British imposed a preference for it, but because the legal infrastructure that would determine whether your contract was enforceable operated only in English. A Cantonese merchant in Hong Kong needed English for the same reason a merchant anywhere needs to speak the language of the court that will hear his disputes: because the alternative is no legal recourse.
Macaulay’s Minute of February 2, 1835, made this logic explicit and extended it. Macaulay proposed redirecting the East India Company’s educational budget away from Arabic and Sanskrit institutions and toward English-medium education. His argument was not aesthetic — his contempt for vernacular scholarly traditions was total and explicit, but it was a bureaucrat’s contempt, not a literary critic’s. His actual argument was efficiency: producing English-literate intermediaries was the optimal use of limited funds, given that all high-value employment in the colonial economy ran through English literacy. He proposed abolishing the Madrassa and Sanskrit College in Calcutta, stripping student stipends from the institutions at Benares and Delhi, and halting the printing of Arabic and Sanskrit texts — redirecting every freed resource into English-medium schooling. He routed the incentive structure deliberately. The dismantling of the existing scholarly infrastructure was not an accident of administration. It was the mechanism.
The consequences did not dissolve with the British Raj. A parent in Mumbai in 2026 who pays a premium for English-medium schooling over Marathi-medium schooling is not expressing cultural deference to Britain. She’s responding to an employment market that still prices English-medium education higher — the same routing, operating 190 years later, without political maintenance.
The embedding repeated across the port cities of British imperial commerce. In Calcutta, Bombay, Madras, Singapore, and Hong Kong, English-language contracts, courts, and accounting standards were not preferences — they were the operating conditions of commerce. In Lagos, Nairobi, and Accra, English common law and English commercial codes became the legal infrastructure of the extractive economy. In each case, the same mechanism: route economic access through the language long enough, and you don’t need to maintain the routing politically. The routing maintains itself because every participant in the existing system has already priced the language into their human capital.
Persian and the Mughal precedent Persian served as the Mughal administrative lingua franca for the same structural reason English became the British one: it was the language of the courts, the contracts, and the bureaucracy that organised economic and political life. When the East India Company displaced Mughal administration, Persian's institutional base collapsed within decades, despite centuries of literary and bureaucratic tradition. This demonstrates, usefully, that English had no special cultural immunity. It won for the same reason Persian won under the Mughals, and lost for the same reason Persian lost when the Mughals went.
1944 and the language of the world’s new infrastructure
The British Empire dissolved across the postwar decades. The incentive structures it had built did not. But that continuity alone doesn’t fully explain what English became in the second half of the twentieth century. Something else happened in 1944.
Before the Second World War, international science was genuinely multilingual. German still held prestige in chemistry and physics. French carried weight in mathematics. Scientists working at the highest level needed multiple languages, and the journals that mattered were spread across them. What ended this was not cultural preference. It was institutional.
The unprecedented expansion of American scientific capacity after 1945 — the National Science Foundation established in 1950, the Cold War research complex pouring federal money into American universities on a scale that had no international equivalent — made American institutions the career destination of the world’s most capable graduate students and American journals the publication venues that determined whether a scientific career advanced. You published in English not because English was the language of better science but because English was the language of the journals with the citation networks that made careers. The mechanism is identical to the East India Company’s: route the incentive through the language, and the language accumulates value it did not earn on its merits.
Two institutional decisions in 1944 deserve particular attention.
In July, forty-four Allied nations gathered at Bretton Woods, New Hampshire, and produced the agreements that created the International Monetary Fund and the World Bank — institutions that would control access to international credit for the following century. They operated in English from their inception. The dollar became the world’s reserve currency; English became the reserve language of international finance by the same mechanism and at the same moment.
In December, fifty-two states signed the Chicago Convention, establishing the International Civil Aviation Organization. English was the convention’s dominant working language, reflecting the dominance of American and British aviation infrastructure in the postwar world. ICAO’s 1951 recommendation formalised English for international aeronautical radiotelephony. Binding proficiency requirements followed in ICAO Annex 1, Amendment 164, adopted in 2003 and implemented in March 2008. Aviation chose English not because pilots were statistically more likely to speak it, but because the institution setting the standards was, from its foundation, one where Anglophone powers had primary influence.
By the 1980s, English was simultaneously the language of global science, aviation, international finance, and the dominant commercial legal standard across former British territories. Not because anyone chose this in a coordinated way. Because each system independently embedded the same language at its founding, and each embedding raised the cost of not speaking English for everyone operating in that system.
The Grin calculation François Grin, an economist at the University of Geneva, estimated in a 2005 report for the French Haut Conseil de l'Évaluation de l'École ("L'enseignement des langues étrangères comme politique publique") that the United Kingdom derives a net annual benefit of €17–18 billion from not having to fund foreign-language education at the scale every other EU country must. The report states that this exceeds "more than triple the famous British Rebate." English dominance is not a neutral equilibrium. It is an ongoing transfer from every non-English-speaking country's educational budget to the Anglophone countries that bear none of those costs. The language's value flows upward — to those who already own it — from the effort of everyone who must acquire it.
The self-reinforcing loop
The historical question — how did English get here? — now gives way to the structural one: why does it stay?
The answer is coordination logic. The value of speaking a language is proportional to the number of people you can reach who also speak it. Each new speaker raises the value for every existing speaker. And because linguistic coordination is total — you cannot run two conversations simultaneously the way you run two software platforms on one machine — once a language achieves critical mass in a domain, the cost of not speaking it exceeds the cost of learning it for every rational actor in that domain. The equilibrium is self-reinforcing: the larger the network, the higher the cost of opting out, which means more people opt in, which makes the network larger.
The wage data illustrates the paradox clearly. Azam, Chin, and Prakash, writing in Economic Development and Cultural Change in 2013, found that fluent English proficiency raises hourly wages for Indian men by approximately 34%, and speaking some English raises them by 13% — returns comparable to completing secondary school. That’s the network premium in a market where English is still a genuine differentiator.
But in European markets where English is already saturated, the picture inverts. Liwiński’s 2019 study in Empirica found that advanced command of English yields a wage premium of roughly 11% among Polish workers — while Spanish commands 32% and French 22%. English’s declining marginal wage premium in advanced economies is not a sign of weakness. It is the signature of saturation. At a certain level of diffusion, English stops being a differentiator and becomes a baseline. Not having it closes off access to the largest total opportunity set in the world. Having it provides no additional edge. That’s not weakness. That’s how universal infrastructure behaves.
The “language of last resort” mechanism compounds this. When a Korean businessperson meets a Brazilian counterpart, they conduct the transaction in English. Neither is comfortable. Neither chose it deliberately. No English-speaking country is involved. English accumulates value from that bilateral interaction — as it does from billions like it — adding to a network no one built for it.
Rakuten’s decision in 2010 wasn’t isolated. Sodexo in France, Nordea in Scandinavia, and Cemex in Mexico have all designated English as their internal corporate working language. They adopted English because it reduces coordination costs across internationally dispersed workforces. The language serves the same function a common technical standard serves: it isn’t necessarily the best protocol in existence, but it’s the one that everybody’s systems already speak.
Why can’t this be displaced the way other dominant standards have been? Because linguistic switching costs are categorically different from technology switching costs.
When a software platform gets displaced, the switching costs are concentrated. A user migrates data, changes habits, learns a new interface. The costs are real, but they’re located in individual decisions that can be made sequentially. Linguistic switching costs are distributed across the educational infrastructure of entire countries — teacher training, curricula, textbooks, examination systems — the professional accreditation systems of every high-status occupation, and the accumulated career capital of hundreds of millions of professionals. These are independent institutional systems with independent governance. Displacing English requires coordinating migration across all of them simultaneously, not sequentially. And simultaneous defection across independent systems doesn’t happen in response to economic pressure alone. It requires a structural rupture.
English currently accounts for 49.6% of web content by language, according to W3Techs data from May 2026. Spanish, the second most common, accounts for 6%. The gap is not narrowing.
The Mandarin ceiling
What would displacement actually require? And is Mandarin producing those conditions?
The surface economic case for Mandarin is real. China is the world’s largest economy by purchasing power parity, a position it has held roughly since 2014–2016 by IMF measures. China is the largest bilateral trading partner for approximately 60 economies, nearly twice the number for which the United States holds that position. The Belt and Road Initiative has built Chinese financial relationships across Central Asia, Africa, and parts of Europe at a scale that superficially resembles British imperial commercial embedding.
The analogy collapses precisely where it matters.
English was embedded when global communication infrastructure was being built. Commercial legal codes, scientific journal systems, aviation standards, international financial institutions — all of these were new and institutionally plastic when English got written into them. Mandarin is entering a system with seventy to eighty years of English-language infrastructure already in place. Replacing a protocol that has been embedded across independent systems doesn’t work like replacing a product in a competitive market. You don’t win users by offering a better alternative. You need to migrate all those systems simultaneously, and they have independent governance.
The scientific publication test makes this concrete. In 2016, China surpassed the United States to become the world’s largest producer of scientific papers by volume, as documented by NSF Science and Engineering Indicators data. Chinese scientific output has grown enormously. Chinese-language scientific publication has not become a global standard. The reason is straightforward: Chinese scientists publishing their highest-impact work publish it in English-language journals, because those journals carry the citation networks that determine career outcomes. The language of publication is determined not by the nationality of the researcher but by the institutional infrastructure of the journals. China changed who produces science. It did not change where science circulates.
Mandarin enrollment in US universities peaked in 2013 and has dropped in every MLA survey period since — down 13.1% between 2013 and 2016, and another 14.3% between 2016 and 2021. During the same period, Korean enrollment rose: up 13.7% between 2013 and 2016, then another 38.3% between 2016 and 2021 — a combined increase of roughly 57% during the peak decade of discussion about China’s economic rise. People making rational individual investments in language learning moved away from Mandarin and toward Korean. The network premium of Mandarin, despite China’s economic scale, does not justify the acquisition cost in the calculations of actual learners.
The acquisition cost is real but secondary. Mandarin is Category V on the FSI scale — approximately 2,200 hours for professional working proficiency, versus 600 for French or Spanish. But a learner in Nairobi choosing between English and Mandarin is choosing between the existing network and the network that would exist if enough other people made the same choice first. Difficulty is not the primary barrier. Coordination is.
What would trigger simultaneous defection? ICAO would have to amend aviation communication standards — a negotiation among its 193 current member states. The scientific journal ecosystem would have to restructure citation networks around Chinese-language publication. The IMF and World Bank would have to change working languages. Multinational corporations across dozens of industries would have to rewrite internal language policies. These systems have independent governance. No single economic fact — including China surpassing the United States in nominal GDP — triggers them simultaneously.
The Latin precedent is instructive. Latin’s displacement took three centuries and required simultaneous structural changes from independent directions: the Protestant Reformation, the vernacular printing press, the rise of national scientific academies, and the collapse of papal institutional authority across Europe. No single event caused it. The coordination equilibrium collapsed because multiple independent institutional systems shifted within a narrow enough window.
The Mandarin optimists are not wrong that economic power matters. They are wrong about how much is needed and what other conditions must be met simultaneously.
Belt and Road and Confucius Institutes: does the embedding mechanism replicate? The British commercial embedding worked by routing legal infrastructure through English: the contracts were in English, the courts enforcing them ran in English, and participation required the language. Does BRI replicate this? In recipient countries, the loans are often denominated in dollars or renminbi, but legal documentation is typically in English. Chinese construction projects do not use Mandarin as the operational language of the supply chain. The Confucius Institute network, which once numbered around 100 in US universities alone, has contracted sharply — as of 2023 GAO findings, fewer than five remained operational in the United States. Where the network has expanded, it has built Mandarin instruction infrastructure, but instruction is not embedding. The EIC didn't teach Indians English; it built a legal system that required English to navigate, which produced demand for English instruction. BRI has not embedded Mandarin in contract law, technical standards bodies, or scientific publication infrastructure in recipient countries. The routing mechanism is not replicating.
The structure we didn’t choose
Approximately 1.5 billion people are currently learning or using English, according to British Council data. Every one of them is spending years of cognitive effort, educational expenditure, and opportunity cost to access a network they did not build and cannot opt out of without professional consequence.
This is, economically, a transfer. Grin’s 2005 estimate put the UK’s annual net benefit from not bearing foreign-language education costs at €17–18 billion — a figure exceeding, in his calculation, three times the value of the British EU budget rebate. Amano and colleagues, writing in PLOS Biology in 2023, documented that non-native English speaking scientists spend up to 91% more time reading scientific literature than their native-speaking counterparts, and up to 51% more time writing journal articles. A career spent at that deficit is not a stylistic inconvenience. It is a structural disadvantage — measurable in euros at the policy level, and in working weeks at the desk.
The mechanism is now self-sustaining. It requires no political maintenance, no ongoing colonial administration, no deliberate exercise of cultural power. Each country that is not natively English-speaking bears the full cost of accessing the network. Anglophone countries bear none of it. The asymmetry compounds silently, without a switch.
No institution is empowered to correct it. The most widely spoken second language in human history runs on a mechanism — commercial embedding, institutional lock-in, network self-reinforcement — that no longer needs the political arrangements that built it, and that costs everyone who didn’t.
Hiroshi Mikitani understood exactly what he was doing in March 2010. He looked at the arithmetic and concluded that the cost of restructuring thousands of professional lives was lower than the cost of operating outside the network. He was right.
Gen AI Disclaimer
Some contents of this page were generated and/or edited with the help of a Generative AI.
Media
Key Sources and References
Mehtabul Azam, Aimee Chin, and Nishith Prakash, “The Returns to English-Language Skills in India,” Economic Development and Cultural Change, Vol. 61, No. 2, pp. 335–367, 2013. https://www.journals.uchicago.edu/doi/abs/10.1086/668277
Tatsuya Amano et al., “The manifold costs of being a non-native English speaker in science,” PLOS Biology, 2023. DOI: 10.1371/journal.pbio.3002184. https://journals.plos.org/plosbiology/article?id=10.1371%2Fjournal.pbio.3002184
François Grin, “L’enseignement des langues étrangères comme politique publique,” Haut Conseil de l’Évaluation de l’École, September 2005. [Report commissioned by the French government; cited for the €17–18 billion annual UK benefit estimate.]
Scott L. Montgomery, Does Science Need a Global Language? English and the Future of Research, University of Chicago Press, 2013. https://press.uchicago.edu/ucp/books/book/chicago/D/bo10984617.html
Thomas Babington Macaulay, Minute on Indian Education, February 2, 1835. Full text available at: https://franpritchett.com/00generallinks/macaulay/txt_minute_education_1835.html
Jakub Liwiński, “The wage premium from foreign language skills,” Empirica, Vol. 46, No. 4, pp. 691–711, November 2019. DOI: 10.1007/s10663-019-09459-0. https://link.springer.com/article/10.1007/s10663-019-09459-0
Modern Language Association, “Enrollments in Languages Other Than English in United States Institutions of Higher Education, Fall 2021,” MLA, November 2023. https://www.mla.org/content/download/191324/file/Enrollments-in-Languages-Other-Than-English-in-US-Institutions-of-Higher-Education-Fall-2021.pdf
W3Techs, “Usage Statistics and Market Share of Content Languages for Websites,” May 2026. https://w3techs.com/technologies/overview/content_language
National Science Foundation / National Science Board, Science and Engineering Indicators 2018 (NSB-2018-1). https://www.nsf.gov/statistics/2018/nsb20181/
International Civil Aviation Organization, “Member States,” ICAO. https://www.icao.int/about-icao/pages/member-states.aspx
US Government Accountability Office, “China: With Nearly All U.S. Confucius Institutes Closed, Some Schools Sought Alternative Language Support,” GAO-24-105981, October 2023. https://www.gao.gov/products/gao-24-105981
Rakuten, Inc., “Englishnization: A personal journey,” Rakuten Today. https://rakuten.today/blog/englishnization-personal-journey.html
British Council, “English in numbers.” https://www.britishcouncil.cn/en/EnglishGreat/numbers
Guzman Manero, “Italian government wants to stop businesses using English — here’s why it’s the lingua franca of firms around the world,” The Conversation, 2023. https://theconversation.com/italian-government-wants-to-stop-businesses-using-english-heres-why-its-the-lingua-franca-of-firms-around-the-world-202312
English 
Español 
Français 
Deutsch 
日本語 
Polski