According to India's 2011 census, only 260 million Indians spoke English—about 22 percent of the population at the time. The 2001 census reported 224 million English speakers, suggesting that the number was not growing especially quickly. Yet English proficiency remains a gateway to many high-paying opportunities. Wealthy Indians conduct much of their business in English, particularly when it depends on modern education or specialized training. Anyone seeking the same opportunities must therefore be able to work in that language.
This divide is easy for affluent Indians to underestimate. Many of us describe ourselves as middle class even when we are wealthy by national standards. The language barrier is consequently not just cultural; it helps determine who can participate in the most productive parts of the economy.
As technology reshapes work, a growing share of economic opportunity will depend on skills and knowledge that are most readily available in English. The Indian information-technology sector illustrates the imbalance: it employs a small fraction of the population while contributing a much larger fraction of GDP. How can people who do not speak English gain access to such opportunities? I see three broad approaches.
Teach English to more people
The slow growth in the number of English speakers between 2001 and 2011 suggests that education alone will not close the gap quickly. People also tend to acquire languages most fluently when they begin young. Even an ambitious English-education program would do little in the near term for the hundreds of millions of adults who are already beyond school age.
Use local languages
Much of India's daily business could be conducted just as well in local languages. We could try to restructure institutions around them. But the approach is difficult to scale when so much scientific, technical, and professional material is produced in English. I am writing this essay in English rather than Hindi or Tamil, which demonstrates the problem.
After independence, India chose to retain English in many of its institutions. Whether that choice was wise is now secondary: it created both advantages and costs that we must address.
Invest in machine translation
Technology sometimes offers a practical route around a deeply rooted social problem. Machine translation may do so here.
Neural translation systems have become remarkably effective for many translations into English. Translation quality in other directions has often lagged, in part because far less training data exists for many Indian languages. Researchers may eventually build equally good systems from smaller datasets, but technological progress is difficult to predict and may take years.
We should invest in large, high-quality multilingual datasets that translation systems can learn from. With enough data, useful translation into Indian languages may be possible with technology that already exists.
Governments, businesses, universities, and nonprofit organizations can help create and curate translated documents. Scaling these efforts could substantially shorten the time required for machine translation to become dependable in everyday life. Human translations are also immediately valuable, even before they become training data.
I am interested in helping this work grow—by connecting relevant organizations, understanding the technical requirements, and estimating how much high-quality data is necessary. Projects outside one's primary research area are also a valuable way to meet people with different experiences and perspectives. My work with MIT EECS GAAP and PathCheck showed me how rewarding that can be.
Originally published February 16, 2022. Statistics are retained from the original essay and refer to the sources and context available at that time.