Candidates preparing for Indian entrance and civil service examinations in Hindi, Tamil, Marathi or Bengali face a real and material disadvantage. Almost all of it sits in the supply of information around the examination rather than in the candidates.
The disadvantage begins with supply. Explanatory material, worked solutions and informed commentary on what a paper actually rewards are produced first in English and translated afterwards, if at all. The lag matters. An analysis of a new paper reaches an English-medium candidate within days and a Hindi-medium candidate months later, thinner and at similar cost. Where translation does happen it is often literal, and technical registers suffer most from literal treatment. A term of art rendered word by word stops working as a term of art, and the candidate is left holding a phrase that appears in no answer key.
The examination compounds this. Major national papers are set in several Indian languages besides English, but the translated paper is a secondary document. Examining bodies commonly state that where a translated question is ambiguous, the English version is treated as final. As an administrative rule that is defensible. As an experience it places the cost of a poor translation on the one person who cannot detect it, since a candidate reading in Tamil has no English text in front of them to check against. Seconds lost to an oddly worded stem are seconds a competitor did not spend.
Then there is reading speed. Decades of psycholinguistic work show that reading in a second language is slower and more effortful than reading in a first, even among fluent readers. Jim Cummins' distinction between conversational fluency and academic language proficiency explains why candidates who converse comfortably in English are still slowed by the compressed register of an examination stem. John Sweller's work on cognitive load supplies the rest: working memory spent decoding a sentence is not available for the physics inside it. On a paper of fixed length that shows up as fewer questions attempted, which is not the same thing as knowing less.
None of this is a difference in capacity, and the mechanisms of durable learning do not depend on the medium. Roediger and Karpicke's 2006 experiments on retrieval practice found that recalling material outperforms rereading it, and the effect belongs to the act of retrieval rather than to the language the retrieval is conducted in. The same holds for distributed practice, where Cepeda and colleagues' 2006 review of spacing effects found the benefit robust across materials and populations. A learner who self-tests in Kannada and returns to the same material at intervals earns what a learner working in English earns.
That carries an uncomfortable implication for the market. The strongest study methods cost nothing and cross languages intact. The weakest, passive rereading of a translated guide, is what a thin vernacular market has most incentive to sell. Candidates outside English-medium coaching are pushed towards the least effective method by the shape of the supply available to them.
The most transferable thing an examination leaves behind is its own record. This platform's archive of JEE Main past papers holds 8,267 authentic questions from 2002 to 2025, and what it yields is stated as proportions. Matrices and Determinants accounts for 8.8 per cent of JEE Main Mathematics since 2016, some 275 questions across the record. Kinematics in one and two dimensions accounts for 6.6 per cent of recent Physics, 180 questions. Coordination Compounds accounts for 8.7 per cent of recent Chemistry, 171 questions. A proportion carries no accent.
Across the three subjects, 117 topics carry enough questions to describe a pattern at all: 48 in Physics, 41 in Chemistry, 28 in Mathematics. The describable part of an examination is finite, and small enough to set out in an afternoon in any language. Part of what a coaching town sells is precisely this sense of proportion, transmitted in conversation, in English or in an English-mixed idiom, inside rooms that cost money and a change of city to enter. Published as evidence in the candidate's own language, it stops being a privilege of geography.
Scoring rules translate perfectly and are still widely misread. The accuracy at which guessing breaks even is the penalty divided by the sum of the reward and the penalty. For UPSC Prelims that figure is 25 per cent, exactly the expected accuracy of a blind pick from four options, so guessing is neutral in expectation and the decision turns on whether a candidate can eliminate even one option. For NEET and JEE Main the break-even figure is 20 per cent, which makes a blind pick from four mildly positive across a full paper. The arithmetic takes two sentences. It circulates mostly in English.
Three things close the measurable part of the gap. The first is structural evidence about the paper, translated early and translated properly, so that the decision about where to spend a final month rests on the same information in every language. The second is bilingual anchoring rather than wholesale translation: reasoning can happen in the home language, but the technical lexicon should stay the one printed on the paper and its answer key. The third is timed practice on the translated paper itself, so that the reading cost is absorbed during preparation rather than discovered in the hall.
The rest belongs to the examining bodies, which control translation quality and could publish parallel texts and corrections as routine. Until they do, the burden sits with preparation. It is worth being exact about which part of the disadvantage can be closed. The language of instruction does not change what a paper asks, how often it asks it, or what a blank answer is worth. Those are facts, and facts do not mind which language they arrive in.
The full set, with progress tracking and five agent perspectives per question, is in the JupiteX app — browse the exam catalogue or browse the Learn library.