Understanding

I. What Does “Understanding” Mean?

When we say “I understand,” what do we mean? That we have decoded the words in a sentence? That we have grasped what the speaker was trying to say? That we have caught what was left unsaid, what was implied, the emotion behind the tone of voice, the centuries of accumulated meaning carried by a cultural reference?

All of these are “understanding,” but none of them are the same thing.

Pragmatic linguistics separates the act of understanding into layers:

First layer – word decoding. Resolving the dictionary meaning of words. “I see” = “I see.” This is the most basic level, and it can be acquired quickly even in a second language.

Second layer – inferential understanding. Extracting what was not said. When “I see” comes from the mouth of a scientist, it may mean “I understand” or “I have noticed.” This requires reading context beyond word knowledge.

Third layer – contextual understanding. Catching cultural and situational nuance. When “I see” falls from the lips of a British diplomat, it may be irony, it may be doubt, it may be a polite refusal. To understand this, you need to know the codes of that culture.

Fourth layer – deep understanding. Accessing the speaker’s worldview, their motivation, what they imply but do not articulate. Reading what a text says alongside what it does not say. Being able to see the structure of an argument, to sense its inconsistencies, to question its underlying assumptions.

Second-language speakers mostly stay at the first layer. Some reach the second. Very few reach the third. The fourth – not everyone reaches it even in their mother tongue.

And perhaps the real problem is this: people say “I understand” while still at the first layer and never question what lies beyond.

But there is another dimension to this layering. The “I see” example itself brings it to light: in English, understanding is built on the metaphor of sight. “I see what you mean,” “that’s clear,” “she shed light on the issue,” “he was in the dark” – as George Lakoff and Mark Johnson (1980) documented in their conceptual metaphor theory, knowing = seeing. This metaphor has burrowed so deep that English speakers no longer even perceive it as metaphor. In Turkish, however, the primary metaphor for understanding is different: kavramak – to grasp with the hand. “I grasped the subject,” “I caught the issue,” “I held the idea.” In Turkish, understanding is built on touch and holding; in English, on sight and light. The same cognitive process, mapped through two different bodily experiences.

This difference is not cosmetic – it reveals the embodied roots of thought. And as the later sections of this essay will show, every language has its own metaphor system, its own conceptual world, and its own “window.” This subject has a continent of its own – this essay goes to the edge of that continent but stops at the shore; the road continues in the next study.

Knowing Terminology Is Not Understanding

This distinction holds across every field. In software engineering, knowing the terms “function,” “class,” and “inheritance” requires English. But “why was this function designed this way?”, “what is the thinking behind this architectural decision?”, “what is the root cause of this bug?” – these are not terminology, they are comprehension. And comprehension happens in the mother tongue.

In law, this is even sharper. Turkish legal concepts like haksız fiil (tortious act), iyiniyet (good faith), muvazaa (simulation), kadastro (land registry) – these are structures that have evolved over centuries within a specific legal order, concentrating that society’s sense of justice, commercial practices, and property relations. As the American Society of Comparative Law puts it, “legal language is a language whose dependence on culture surpasses all other specialist languages.” When you translate a concept into another language, you transfer the word; you cannot transfer the institutional and cultural deposit beneath it.

The Turkish legal concept of vicdan (conscience) makes this concrete. Vicdan derives from Arabic wijdān but has acquired a different density of meaning in Turkish over the centuries. Vicdani kanaat (a judge’s conviction of conscience), vicdanın sesi (the voice of conscience), vicdan azabı (torment of conscience) – these can be translated into English as “conscience,” but conscience in English is an individual moral compass. Turkish vicdan carries both an individual and a communal dimension: a judge’s vicdani kanaat is not purely individual but encompasses a collective sense of justice. When you translate this concept into English, you transfer the structure of the word but cannot transfer the cultural weight beneath it – just as muvazaa can only be captured at the dictionary level by translating it as “simulation.”

In medicine, anatomical terms are universal – a femur is a femur everywhere. But clinical reasoning, patient communication, and ethical decision-making are language-dependent. In a survey in Saudi Arabia, medical students trained in English preferred English instruction in nearly every part of the curriculum, while only in communication skills did a notable minority – 28% (131 of 468 students) – want that course taught in their mother tongue, Arabic (Alrajhi et al., 2019).

The same structure recurs in every field: terminology may be international, but expression and comprehension belong to the mother tongue. Any education system that ignores this distinction mistakes the first layer for the fourth – it thinks it has taught the concept because it has taught the terms.

II. Thinking in a Second Language: What Does Science Say?

The Brain Works Differently in a Foreign Language

In 2012, Boaz Keysar and colleagues at the University of Chicago published one of the most striking findings in cognitive psychology. It is called the “Foreign Language Effect.”

The experimental scenario: an epidemic will kill 600 people. Two programmes are offered.

First frame – in the language of gains: “Programme A will definitely save 200 people. Programme B has a 1/3 chance of saving all 600 and a 2/3 chance of saving no one.” People choose A – they take the sure gain and avoid risk.

Second frame – in the language of losses: the same scenario, different words. “Under Programme A, 400 people will definitely die. Under Programme B, there is a 1/3 chance no one dies and a 2/3 chance all 600 die.” This time people choose B – they reject the certain loss and take the risk.

The two frames are saying exactly the same thing mathematically. But in their native language, people decide differently – because “save” and “die” carry different emotional weight. This is Kahneman and Tversky’s Nobel Prize-winning framing effect.

Keysar’s discovery: in a foreign language, this asymmetry disappeared. People made the same decision under both frames – consistent, systematic, unswayed by emotional framing. The words “save” and “die” lost their emotional weight in the foreign language. Same words, different burden.

The mechanism: a foreign language provides “greater cognitive and emotional distance.” The mother tongue is loaded with emotional experience from childhood onward – the language of a first love, the language of a mother’s scolding, the language of childhood fears. “Death” in the mother tongue is not merely a concept; it is the sound of funerals, of lost relatives, of childhood terrors. In a foreign language, the same word – “death” – remains a dictionary entry. Meaning is the same; weight is different.

What does this mean? Thinking in a foreign language is qualitatively different from thinking in the mother tongue. Same person, same problem, two different languages – two different decisions. And at first glance the decision made in the foreign language looks “more rational” – but that rationality is the product of lost emotional depth. Emotional context is not always “bias”; sometimes it carries information. When a judge pronounces the word “killing” without the slightest tremor, is that rationality, or a sign that the sense of justice has gone numb?

Hayakawa and Keysar (2018) took this further: using a foreign language even weakens mental imagery. When people imagine a scene in a foreign language, the mental picture is less vivid. Even daydreaming is deeper in the mother tongue.

And there is neuroscientific evidence. Szczypek and colleagues (2026) measured event-related brain potentials (ERP) in Polish-English bilinguals. Finding: the brain’s response (N400) to moral violations in the second language was reduced compared to the first language. The brain processes a moral violation in the foreign language “more superficially” – it understands the word but cannot fully access the meaning network beneath it.

Circi and colleagues’ 2021 meta-analysis covers 17 studies and 47 experiments. Finding: the foreign language effect is consistently replicated, and it is stronger between distant languages (such as Turkish and English).

Inner Speech: What Language Do You Think In?

“Can you think in English?” sounds like an abstract question. But in cognitive science it has a concrete counterpart: inner speech. That voice in your head – the one you hear when planning, deciding, arguing with yourself – what language is it speaking?

Jean-Marc Dewaele (2015) conducted a large-scale study with 1,454 adult multilinguals. Results:

• In general inner speech, the mother tongue dominates. People think predominantly in the first language they learned.
• In emotional inner speech, the mother tongue is virtually the only language: anger, joy, sadness, fear – all experienced in the first language.
• Inner speech in the second language increases only with long-term residence in that language’s environment – that is, with acculturation.

Hammer (2019) examined 149 highly educated, second-language proficient, acculturated bilinguals. Finding: the second language can be used for work and academic thinking, but emotion, dreams, and automatic responses stay in the mother tongue. And the full transition is never complete – inner speech oscillates between the two languages.

Ardila and Rosselli’s (2017) finding adds a striking detail: bilingual people always do mental arithmetic in the first language they learned. Even at advanced second-language proficiency, the answer to 7 times 8 comes from the voice of the first language.

Resnik (2021) concludes: the language shift in inner speech is a process that takes years, and in most people it never completes. A few years of university education are not enough to change automatic inner speech.

BICS and CALP: Two Different Kinds of “Knowing”

Jim Cummins introduced one of language education’s most influential conceptual distinctions in 1979:

BICS (Basic Interpersonal Communication Skills): Everyday conversation, greetings, simple communication. Acquired in 1–3 years.

CALP (Cognitive Academic Language Proficiency): Abstract thinking, expressing complex concepts, constructing arguments, analysis, synthesis. Requires 5–10 years.

Why is this distinction critical? Because Turkey’s preparatory-year model primarily builds proficiency at the BICS level. In one year, students acquire conversational skills and appear “fluent” – but university education requires CALP-level proficiency. There is a 4–9 year chasm between the two. But CALP is not the final destination either. CALP is the threshold for abstract thinking in an academic language – using terms correctly, constructing arguments, performing analysis. But walking the road behind the terms is something else entirely. To define tort is CALP. But the person who defines tort simultaneously does not truly know tort – because the definition is not the concept. The definition is a shortcut drawn to reach the student; the concept is far larger than the definition. A tort comes to life not in a definition but in an event. Applying that concept in a real dispute requires having walked the road that leads to it – having internalised hundreds of cases, centuries of legal evolution, the subtleties of a society’s sense of justice. This is not a skill measurable in units of time but a depth formed through lived experience. A definition is a map – with a map you find your bearing and destination, but you cannot live the road. The brain struggles to fully walk this road even in its own mother tongue. To then say “now walk it in a foreign language too” is to mistake the map for the road.

And this chasm creates an illusion: the student appears to “know English,” the teacher and the system act accordingly – but in reality the student is decoding words without reaching the depths of thought. The system mistakes BICS for CALP; those who reach CALP believe they have reached the world behind the terms. The same illusion at every layer: the place where you stand looks like the final stop. Khatib and Taie (2016) document this directly: the rapid acquisition of BICS creates a false perception of competence, and because of this illusion, the support the student needs is withdrawn.

III. The Illusion of Competence

Shallowness Mistaken for Mastery

There is a paradox here: language knowledge that appears competent can actually reduce the rate of comprehension.

This sounds absurd. How can knowing something be worse than not knowing it at all? But we are not talking about the person who knows little. The shallowness in question is not that of someone whose English stalled at secondary school – it is the shallowness of someone educated in a foreign language, holding certificates, watching television without subtitles, believing themselves proficient. And it is precisely this perception of proficiency that creates the danger. The mechanism works as follows:

If you do not know the language: You know that you do not understand. You seek a translator, you read the context, you consult an expert, you verify through other channels. Not knowing the language makes you cautious. It is entirely possible for someone who does not know English to catch a bad translation through logic, context, and semantic consistency – because a person aware of their own lack of understanding develops compensatory mechanisms.

If you appear proficient but know shallowly: You believe you understand. You decode the words, catch the surface meaning of the sentence, and say “got it.” But you miss the nuance, the irony, the cultural layer, the implication – and you do not know what you have missed. The perception of competence generates false confidence.

In cognitive psychology this is documented as the “illusion of competence.” What Koriat and Bjork (2005) demonstrated is that there is a difference between recognising something and truly knowing it, and people constantly confuse the two. They believe they know what they merely recognise.

Bardovi-Harlig and Dörnyei’s (1998) classic study shows this illusion’s counterpart in the language domain: second-language learners are sensitive to grammatical errors but nearly blind to pragmatic errors. They immediately spot a grammatical mistake in a sentence, but they cannot see that being excessively direct in a request looks rude. They see what is wrong; they do not see what is missing – because they do not know what is supposed to be there.

The Pragmatic Chasm

Lee and Lee (2022) showed that second-language speakers are significantly behind native speakers in comprehending aural sarcasm. Tonal and contextual cues are less effective for second-language speakers.

Chang (2010) examined the apology strategies of highly proficient (C1) Chinese learners of English. Grammar had improved, but pragmatic strategies still carried mother-tongue transfer. A C1 speaker can construct a grammatically flawless apology, but may misjudge the cultural weight, timing, and dosage of that apology.

Cieślicka’s (2015, 2017) finding reveals the mechanics: native speakers process idioms holistically – when they hear “kick the bucket,” they jump directly to the meaning “die.” Second-language speakers process them compositionally – first they decode “kick” and “bucket,” then they recognise it as an idiom, then they work out the meaning. They “understand” the same words but they do not experience the same thing.

Karpava (2025) explains the structural reason for this chasm: native speakers acquire pragmatic competence subconsciously – through thousands of social interactions from childhood. Second-language speakers must consciously learn pragmatic competence. And that learning happens not through education but through living.

A Turkish-English Finding

Sıtkı, Ikier, and Şener (2024) worked directly with Turkish-English bilinguals. Finding: false memory in the second language is lower than in the mother tongue. The mechanism: the connection between the word store and the semantic system is weaker in the second language.

What does this mean? You know a word in the second language, but you do not know the associative network it carries, its emotional charge, its cultural layers. “Understanding” in the mother tongue is a rich semantic network activation – in the second language it is a narrow channel. Same word, two different depths.

IV. What Is the System Doing?

Why Do Universities Teach in English?

The answer to this question has nothing to do with educational quality.

The British Council’s (2017) comprehensive research is explicit: the reasons universities adopt English-Medium Instruction (EMI) are as follows:

• Attracting international students (revenue)
• Rising in global rankings
• Projecting an “international” image
• Attracting international faculty

The British Council sums it up: “The decision to use English as a medium of instruction in higher education is a management decision aimed at raising the institution’s profile rather than a desire to experiment with a new language teaching approach.”

In other words, English-medium instruction is chosen not so the student thinks better, but so the university looks better.

4 Years = 10 Weeks

The most striking piece of data from the same British Council research: over four years of education at English-medium universities, students’ average IELTS improvement is 0.5 points. An intensive language course achieves the same improvement in 10 weeks.

4 years = 10 weeks.

This datum alone suffices to ask: if the goal is to teach English, why are you spending four years on it? The answer: the goal is not to teach English.

The System Acknowledges the Barrier

Here is the real contradiction. The same system that defends English-medium instruction simultaneously makes accommodations that acknowledge the language barrier’s existence:

Examination practices:

• US universities: one additional hour per exam for non-native English speakers, not exceeding 5 hours total (American University policy).
• ACT exam: up to 50% additional time for non-native speakers. Stated reason: “language processing barriers, reduced reading comprehension, and time spent translating words into the native language.”
• Microsoft/Pearson Vue certification exams: 30 extra minutes for non-native English speakers.
• Universities: separate admissions tracks for international students, reduced course loads, language preparation courses – all standard practice.

Boise State University’s own policy document contains a telling admission: “Writing and reading in a second language takes significantly longer than in a first language, and exam performance may reflect language processing time rather than subject knowledge.”

The system says “English-medium instruction works” on one hand, while giving extra time on exams with the other – because it knows it does not work. But it does not question the model. Because the motivation is not the student’s learning but the institution’s functioning.

The Cost of Science’s Language Barrier

Tatsuya Amano and colleagues’ research published in PLOS Biology (2023) measures this contradiction’s counterpart in the world of science:

• Non-native English-speaking scientists spend approximately twice as long reading an English-language paper as native speakers.
• For a doctoral student, this amounts to 19 extra working days per year – just for reading papers.
• Journals reject papers by non-native English-speaking authors 2.6 times more often on the basis of language quality.
• Non-fluent English speakers receive 12.5 times more revision requests.
• In a separate study screening 736 journals (Arenas-Castro et al., 2024), only 2 explicitly state that papers will not be rejected on the basis of language quality alone.

A sentence from Nature’s 2025 follow-up article sums it up: “Not being able to speak fluent English often means being regarded as a lower-quality scientist.”

There is a language tax at work here – loaded onto the shoulders of the rest of the world, levied upon science, invisible yet merciless. What is being measured is not science but language. What is being evaluated is not thought but performance.

A Shallow Definition of “Success”

All of these accommodations – extra time, language support, adapted exams – expose one thing: the system measures “success” shallowly.

Passed = understood.
Graduated = competent.
Sat the exam = was assessed.

But no one asks: how deeply did they understand? The student who received extra time answered the same question, but did they answer it at the same depth? How would they have answered it in their mother tongue? Which nuance did they cut short in the foreign language? Which argument did they want to construct but could not?

These questions are not measured. What is not measured is invisible. And what is invisible is treated as though it does not exist.

The Law “Exception” Fallacy

Nearly all state university law faculties in Turkey teach in Turkish. When Boğaziçi University opened a law faculty in 2022 – at a university where nearly every other department is in English – it opened a bilingual law programme whose core professional courses are in Turkish. Its own website states the reason explicitly: “In line with global practice, there is no university in our country offering a fully foreign-language law education.”

This is not a Turkish exception. The global picture:

Maastricht University in the Netherlands offers a fully English-language undergraduate law programme – technically an exception. But with a critical detail: this programme teaches not national Dutch law but comparative European law, and it does not grant graduates direct qualification to practise law.

The European Union itself is the largest-scale proof of this reality. The EU must prepare every piece of legislation it enacts in 24 official languages – 552 possible translation combinations. And the legal status of all 24 language versions is equal. The German text is not “more valid” than the French text. Yet the EU Court of Justice regularly faces interpretive cases arising from differences in meaning between language versions. Even the world’s most sophisticated multilateral legal structure cannot reduce legal text to a single language.

Is this only law’s problem? No.

Even in medical education – a field expected to be a universal science – 55.6% of more than 2,900 medical faculties across 189 countries (105 countries) teach in the national language. In Europe, this figure is 97.4% (Hamad, 2023). In Japan, the vast majority of university engineering education is in Japanese. Including the University of Tokyo and Kyoto University. With this education, Japan files 414,413 patent applications per year – third in the world (WIPO, 2024).

Law is not an exception; it is the most visible instance of the rule. Every thought-intensive field – law, medicine, philosophy, engineering – is language-dependent. Law displays this dependence most clearly because legal concepts are inherently local: they cannot be equivalently translated into another legal order. But the mechanism is the same: terminology may be international; comprehension and reasoning belong to the mother tongue.

V. Whose Loss?

Two Profiles, Two Outcomes

The impact of foreign-language education is not the same for everyone. Two different profiles produce two different outcomes:

For the average profile, the system “works.” BICS-level English is acquired, “English – advanced” is written on the CV, a few sentences are strung together in a job interview, a career advantage is gained. No one asks “how deeply do you understand?” because the system already measures shallowly. Grade, diploma, certificate – all in order. There is a marginal boost, and this boost is deemed sufficient.

For the deep-thinking profile, the system can be harmful. And this harm operates through three mechanisms:

First – thought suppression. A person with high thinking capacity can construct complex arguments in their mother tongue, express nuance, generate metaphors, draw fine distinctions. When they switch to a foreign language, all of this drops to the BICS level. A line from a Chinese engineering professor documented by Zheng and Choi (2024) makes this concrete: “I understand the knowledge. But I can’t improvise in English.” The next professor summarised their first year in a single word: panic.

These people’s thinking capacity is not low – the language is preventing them from using their capacity.

Second – energy drain. Reaching CALP level takes 5–10 years. A Turkish engineering student faces roughly these time demands: approximately 1,000 hours of English instruction in the preparatory year, plus an estimated 2,000–3,000 additional hours grappling with English over four years of undergraduate study. That is 3,000–4,000 hours of opportunity cost. And throughout those years, their CALP in their mother tongue is not developing either – because attention, energy, and time are going to the foreign language. The result: intermediate in both languages.

Third – potential pruned. Civan and Coşkun (2016) investigated the effect of instruction language on academic performance at Turkish universities. When entry scores were controlled, the performance of students in English-medium programmes was lower than those in Turkish-medium programmes. Strong input → weak output.

This makes no difference for the shallow profile, because the loss is invisible. But for the deep profile, the loss is enormous – because the potential was large. And this loss is never measured: what that student could have been, could have produced, could have said in their mother tongue – there is no answer to this.

Bälter and colleagues’ randomised controlled study in Sweden (2024) puts numbers to this. Same course, same exam – one group given in Swedish, the other in English. The group taking the exam in Swedish correctly answered 73% more questions than the English group. That is, if a student in the English group answered 4 out of 10 questions correctly, a student in the native-language group answered 7 – two groups taking the same course, learning the same material.

And the most striking detail: students from the English group who rated themselves as proficient in English were examined separately. These were students who said “I know English,” who were confident. They still performed 41% worse than those who took the exam in their native language. The numerical equivalent of the illusion of competence: you say “I know” but you lose 41% – and you are not aware of what you have lost.

Cognitive load theory (Sweller, Roussel and Tricot, 2022) explains the mechanism: the disadvantages of learning in a second language can outweigh the advantages. The brain is forced to split its limited resources between language processing and content processing – and it sacrifices one for the other.

At this point, a truth bears repeating: cognitive energy is a finite resource. No matter how intelligent, how hardworking, how motivated you are – the amount of information your brain can process simultaneously is limited. Glucose consumption, attentional capacity, working memory – all have ceiling values. This is a biological constraint of the human organism. And how a finite resource is allocated is a strategic decision. If energy goes to language decoding, it does not go to comprehension. If it goes to syntax, it does not go to analysis. If it goes to translation, it does not go to original thought. Bälter’s 41% is the concrete invoice of that allocation decision – and the vast majority of those paying this invoice are not even aware they are paying it.

From the Field: Measuring the Depth of Meaning

As part of this research, individuals of different profiles speaking English at international panels were systematically observed. Davos 2025 World Economic Forum, the Geneva Human Rights Summit, AI safety panels – dozens of hours of recordings. The question was simple: how deeply can speakers express meaning?

The picture that emerged points to a variable independent of instruction language:

Two striking comparisons:

Bengio and Lagarde share the same mother tongue: French. Lagarde’s English is more fluent, more polished, less accented. But Bengio’s depth of meaning is markedly greater. In AI discussions, his “foggy mountain road” metaphor, his “baby tiger” analogy, his “rolling the dice with humanity’s future” framework – these are conceptual architecture, not cliché. Same language, different depth → the differentiating factor is not fluency but thinking capacity.

Mengü is one of the most successful products of Turkey’s English-language education system. TED College, Bilkent Political Science, international journalism experience. Her English at the 2023 Geneva Human Rights Summit was fluent, her terminology correct: “competitive authoritarian regimes,” “personality cult,” “checks and balances.” But she did not reconceptualise a single concept from her own perspective – she applied labels without opening them. Historical reference was virtually absent. She introduced Michael McFaul as a “former Defence Secretary” – McFaul is a former US Ambassador to Russia and Stanford professor. A sign that the ideas she had borrowed had not been fully digested.

And Fink – the native English-speaking CEO of BlackRock – sits fifth in the table. Depth of meaning cannot be explained by native-language advantage.

These observations are not statistical evidence; they are systematic observation. But they are consistent with the thesis: depth of meaning depends neither on the language nor on fluency – it depends on thinking capacity. Language is the vehicle that carries thought. No matter how good the vehicle, if there is no thought to carry, it runs empty.

The Honesty of “I Don’t Know”

It cannot be said with certainty that foreign-language education destroys the potential of deep thinkers. This is a hypothesis – a strong hypothesis, but not a proven verdict.

The available data points to this: foreign-language education marginally elevates individuals with low thinking capacity while potentially pruning the potential of those with high thinking capacity. The system measures the first group’s “success” and does not see the second group’s loss – because there is no metric to measure the loss.

The magnitude of this loss is unknown. But there are strong indicators of its existence.

VI. Every Language Is a World

Language Is a Window

Everything described so far – the foreign language effect, inner speech, the illusion of competence, the system’s contradictions – revolves around a single question: how does language affect thought? But this question has a reverse side: how does language display the world? Every line in this section is a research topic in its own right. Here we are merely placing road signs – the road itself is far longer.

In the 1930s, linguists Edward Sapir and Benjamin Lee Whorf advanced a radical hypothesis: the language we speak determines how we think. The “strong” version of this hypothesis – that language completely determines thought – has been largely refuted. People can think without language; infants solve problems before language; deaf individuals engage in complex reasoning without sign language.

But the “weak” version – that language influences thought, directs perception, shapes patterns of attention – has been supported by robust experimental evidence over the past twenty years. And this evidence shows that every language truly does open a different window onto the world.

Two Names for Blue

English has words for shades of blue: navy, azure, cobalt, teal, indigo. But they are all subcategories of “blue” – the parent term is always “blue.” In Russian, the situation is different: siniy (dark blue) and goluboy (light blue) are two separate basic colour terms – not shades of “blue” but colours as independent from each other as red and orange.

In 2007, Jonathan Winawer and colleagues tested whether this difference affects perception. In the experiment, participants were shown three blue squares and asked which was different from the other two. Result: Russian speakers were faster at distinguishing when the two colours fell into different categories (one siniy, the other goluboy). When the same physical colour difference fell within a single category (both siniy or both goluboy), this speed advantage vanished.

Furthermore: this advantage disappeared when participants were given a verbal task (silently repeating an eight-digit number) but persisted when given a spatial task (holding a pattern in a 4×4 grid). The effect genuinely operates through language processing.

Same sky, same wavelength, same retina. But a Russian speaker sees differently from an English speaker – because their language concentrates their attention where it divides the colour space differently.

And this effect is not specific to English. Emre Özgen and Ian Davies’ (1998) research identified 12 basic colour terms in Turkish – against English’s 11. The twelfth: lacivert (navy blue). The interesting point: most participants defined lacivert as “a type of blue” – but at the perceptual level, lacivert operates as an independent category. Turkish speakers are significantly more successful at separating blue tones into two distinct groups compared to English speakers. Metalinguistic judgement – what we say about language – and the language’s actual mode of operation are different things. People say “lacivert is a type of blue” but their brains process them as two separate colours.

This metalinguistic-perceptual split mirrors the essay’s core question: people say “I know English” but their cognitive processes do not confirm it. The surface judgement and the actual mode of operation diverge everywhere.

Three languages, three different maps of blue: English gathers everything under a single “blue”; Turkish separates mavi and lacivert but does not openly acknowledge it; Russian treats siniy and goluboy as independent as red and orange. Every language is strong in a different place – and none of their maps is “correct”; each sharpens a different slice of reality.

The Life Behind the Colour

But where do these differences come from? Why did Russian split blue in two while English did not? Why did Turkish distinguish lacivert but take pink – a colour that the Himba people of Namibia, for example, group under the same term as red – for granted?

The answer reminds us that language is not an abstract system: language is the sediment of lived experience.

Russian’s siniy/goluboy distinction dates back to Old Slavic. In the Slavs’ environment, the light blue of the sky (pigeon-coloured – goluboy, from голубь, pigeon) and the dark blue of deep water (siniy, with connotations of shimmer and depth) were different experiences. Different experience produced different words; different words sharpened perception; sharpened perception reshaped experience. A cycle.

This cycle operates far more visibly elsewhere: Islam and green. In the desert, green is rare. What is rare means life – water, oasis, shade, vitality. This lived experience entered language: in Arabic, paradise (جَنّة, Jannah) means garden. Descriptions of paradise in the Quran are built upon green gardens, flowing waters, shade-giving trees. Green travelled from lived experience into language, from language into the sacred, from the sacred into perception. Today, the colour green carries an automatic connotation of holiness in the Islamic world – this connotation is not biological but the crystallised form of lived experience transmitted through language.

The counter-example confirms this: in the Amazon rainforests, green is everywhere. Not rare, but ordinary. The same mechanism (lived experience → language → meaning) operates here but produces a different result – green is not invested with holiness because green is not the exception of life but life itself.

A historical detail shows how deep this cycle runs: in pre-Islamic Arabic, the semantic range of akhḍar (green) also encompassed darkness and blackness. After Islam, the same word acquired an entirely different density of meaning. Lived experience transformed language; language transformed perception; perception reshaped lived experience.

Being aware of this cycle breaks the illusion that languages differ “at random.” Every language’s map carries the trace of thousands of years of its speakers’ lived experience. If Russian split blue in two, it did so because a Slavic peasant lived the sky and the river water as two separate experiences every day. If Turkish set lacivert apart, it did so because in Anatolia dark blue and light blue pointed to different lives – night and day, the depth of the sea and the sky above.

No language divides the world “as it is.” Every language divides the world as its speakers have lived it.

The Direction of Time

Which way does time flow? Left to right? Top to bottom? East to west?

The answer depends on what language you speak.

English speakers arrange time left to right – yesterday to the left, tomorrow to the right. Arabic speakers arrange it right to left – following the direction of writing.

In Mandarin Chinese, time metaphors are predominantly vertical: “the month above” = last month, “the month below” = next month. According to Lera Boroditsky’s (2001) finding, Mandarin speakers are faster at vertical time arrangement than English speakers. That said, the replication of this effect is contested: some studies failed to reproduce it (January and Kako, 2007), while others re-established it with new methods (Boroditsky, Fuhrman and McCormick, 2011).

But the most striking example comes from Australia’s Kuuk Thaayorre people. In this language, there are no relative directional expressions like left-right or front-back. Everything is expressed in absolute directions – north, south, east, west. They say “the ant on your south foot,” not “the ant on your left foot.” Kuuk Thaayorre speakers maintain compass-like directional awareness at all times – because their language demands it.

Boroditsky and Gaby (2010) gave these people cards – a person ageing, a banana being eaten, a crocodile growing. They were asked to arrange them in order. Regardless of which direction they were seated facing, they arranged the cards east to west – following the sun’s path across the sky. Someone facing south arranged them left to right; someone facing north arranged them right to left; someone facing east arranged them toward their body.

Same universe. Same time. Three different maps – because three different languages, three different windows.

Encoding the Source of Knowledge: Turkish’s Epistemic Advantage

Turkish possesses a feature rare among the world’s languages: evidential markers – the grammatical obligation to encode the source of one’s knowledge.

An example:

“Ali geldi.” – Ali came. I saw it. I was there. I am a direct witness.

“Ali gelmiş.” – Ali came (apparently/reportedly). Someone told me. Or I saw evidence of his arrival (his shoes are at the door) but did not see him myself.

In English, this distinction is not obligatory. “Ali came” covers both situations. If you want to indicate the source of the knowledge, you must add extra words: “I saw Ali come” or “I heard that Ali came.” It can be expressed, but the language does not compel it – you have the option of not saying. In Turkish, this information is embedded in the verb conjugation – you cannot avoid saying it.

Ayhan Aksu-Koç and colleagues’ decades-long research programme has documented the cognitive effects of this grammatical obligation. Findings:

• Turkish children begin using evidential markers from age 2 – but they consistently distinguish between direct experience (-dı) and indirect knowledge (-miş) by age 4.
• This process runs in parallel with the development of epistemic awareness: the ability to ask “how do I know this?”
• Aksu-Koç, Ögel-Balaban, and Alp’s finding is striking: there is evidence that exposure to evidential markers may enhance children’s performance on “false belief tasks.” Turkish grammar may be building the concept of “what someone else knows can be different from what I know” earlier and more robustly.

What does this mean? A Turkish-speaking child, with every sentence they construct, is compelled to classify the source of their knowledge. Did they see it, hear it, infer it? This obligation embeds the habit of epistemic thinking – thinking about the nature of knowledge – within the language itself.

An English-speaking child has no such obligation. They say “Ali came” and move on. The source of the knowledge is left to questioning, probing, noticing – but it is not grammatically enforced.

This does not mean Turkish is a “better” language than English. It means it opens a different window – a window where the source of knowledge is visible.

A World Without Number

The language of the Pirahã people, living deep in the Amazon forests, constitutes one of linguistics’ most contested cases. Linguist Daniel Everett, after over 30 years of fieldwork, reported: the Pirahã language has no number words. No “one,” “two,” “three” – only vague quantity expressions meaning “few” and “many.”

The Pirahã people, worried about being cheated in trade with merchants, asked Everett to teach them numbers. They worked enthusiastically every day for eight months. The result: they could not learn them. After eight months, the Pirahã concluded that the lessons were not working and stopped.

This is not a deficit of intelligence – it is the boundary of a cognitive framework created jointly by language and culture. Everett calls this the “immediacy of experience principle”: Pirahã culture restricts communication to directly experienceable subjects. The abstract concept of number falls outside this framework.

This case is one of the most extreme examples showing that language does not “determine” thought but powerfully frames it. If a language does not encode a particular concept, thinking with that concept becomes difficult – not impossible, but very hard.

The Worldview Inside the Characters

The Chinese writing system is perhaps the most visual example of how language carries a worldview. Each character has engraved the historical and cultural interpretation of a concept into the writing itself.

好 (hǎo) = “good.” The character’s structure: 女 (woman) + 子 (child). Woman and child side by side = good.

安 (ān) = “peace, safety.” Structure: 宀 (roof) + 女 (woman). A woman under a roof = peace.

姦 (jiān; simplified form 奸) = “treachery, adultery, wickedness.” Structure: 女 + 女 + 女 – three women side by side = evil. The 1989 edition of the Cihai dictionary contains 275 characters bearing the 女 (woman) radical. The male radical? None.

These structures have etched a millennia-old patriarchal social order into the DNA of the writing. Learning Chinese is not merely learning an alphabet and grammar – it is entering the historical worldview carried by those characters. In a language where “peace” is encoded as “a woman under a roof,” the concept of peace itself carries a different cultural weight.

Zhou’s (2025) research examines how modern Chinese feminists are problematising these character structures and proposing language reform. The characters are not merely communication tools – the societal assumptions they contain seep into the subconscious of everyone who speaks the language.

Li’s (2016) study strikes a careful balance: rather than summarily declaring “Chinese is a sexist language,” it argues for understanding the historical context of the characters and reading them alongside modern interpretations.

Both positions point to the same conclusion: language is not a neutral tool; it is a carrier of worldview. “Learning” Chinese characters is entering that worldview.

The Hidden Philosophy of Turkish

Every language has untranslatable words. But Turkish’s untranslatables cluster especially densely in the domains of emotion and relationship – each carrying in a single word a world that would require a paragraph to explain in another language.

Gönül. In English one says “heart,” but gönül is not the heart. It is the inner centre where feeling, intention, and devotion meet. “My gönül does not consent” – not my heart, not my mind, something somewhere is refusing but I cannot quite say where. “To place gönül” – a kind of hurt, but not anger; something quieter, deeper. No single English word corresponds, because English’s emotional map divides this territory along different lines.

Hüzün. Orhan Pamuk introduced this to the world in Istanbul: Memories and the City. Melancholy? No – melancholy is individual; hüzün is collective. In Pamuk’s formulation: “A dark mood shared not by one person but by millions.” A collective sense of loss for Istanbul’s past, the fall of the Ottoman Empire, vanished grandeur – yet not hopeless; in Sufi philosophy, hüzün is feeling one’s distance from God and finding beauty in the ache of that distance.

Keyif. English “pleasure”? No – pleasure has an object, enjoyment derived from something. Keyif has no object. Keyif is a quiet contentment arising from the act of existing. “Coffee keyif” – the coffee is an occasion, but the keyif does not come from the coffee; it comes from pausing, from allowing time to pass, from the absence of pressure to produce anything. English has “leisure,” “relaxation,” “rest” – each implying recovery from something. Keyif is not recovering from anything. It simply exists on its own.

Merak. English “curiosity”? Partly. But in Turkish, merak also means “worry.” “I merak you” can mean “I am very curious about you” or “I am anxious about you.” Merak is both the desire to know and concern – Turkish sees these two feelings as kin. In English, there is no kinship between “curiosity” and “worry.”

Gurbet. “Living in gurbet” – translated into English as “living abroad,” but gurbet is not “abroad.” It is the fusion of spatial distance from home and spiritual disconnection. A gurbetci is not simply someone who is far away; they are someone who carries the distance inside them. “Diaspora” comes close, but diaspora is political; gurbet is existential.

Hasret. “Longing”? “Yearning”? Neither carries the weight of hasret. It is the chronicled, heavied, nearly physicalised form of missing. Özlem (another Turkish word for missing) is more momentary and personal; hasret is longer and deeper. Turkish draws an intensity gradient between özlem and hasret – English has no such gradient.

Emek. “Labour”? “Effort”? Neither captures emek. It is work and self-sacrifice and love combined. “To give emek” – not just to work, but to pour something of yourself into it. “Health to your emek” – health not to your hands but to your being.

These concepts are not merely “beautiful untranslatable words” – they concern the resolution of thought. English says “understand” – one word, the same at every depth. Turkish has: anlamak, kavramak, idrak etmek, sezmek, fark etmek, çözmek – each pointing to a different mental operation. Kavramak is grasping with the hand, sezmek is intuiting, idrak etmek is a cognitive leap. English says “understand” for all of them and moves on.

The same holds for emotion. English has adore, cherish, be fond of, but in daily speech “love” stands in for all of them. Turkish: sevmek, âşık olmak, tutulmak, gönül vermek, bağlanmak, düşkün olmak – each carries a different relational tone. And in Turkish, you cannot use the same verb for pizza and a person – you can sevmek pizza but you cannot fall in love with an object. English uses “I love pizza” and “I love you” with the same verb – is that flexibility, or the dilution of feeling?

The more words a language has to distinguish an experience, the more precisely it can think about that experience. English’s habit of loading the same word onto ten different contexts looks like vocabulary breadth, but it has a cost: meaning blurs. Turkish does the opposite – it produces its own word for each situation. This is not a luxury; it is the resolution of thought.

These words do not show that Turkish is a “better” language. Every language has its own untranslatables:

Portuguese’s saudade – a deep longing for the past but with a dimension reaching into the future; the concept of “future saudade” (saudades do futuro) expresses missing things that have not yet been experienced.

Japanese’s mono no aware – a melancholic sensitivity to the beauty of transience; the feeling experienced while watching cherry blossoms fall; it is impermanence, not permanence, that is beautiful.

Japanese’s wabi-sabi – the beauty in imperfection, incompleteness, transience; the aesthetics found in a cracked tea bowl.

Japanese’s komorebi – the dancing pattern of light spots on the ground created by sunlight filtering through leaves. A single word.

Finnish’s sisu – a fusion of willpower, resilience, and courage; “the second wind that emerges when things get tough” but more than all of these.

German’s Schadenfreude – pleasure taken from another’s misfortune; English has no name for this feeling (the word was borrowed directly from German).

Danish’s hygge – a feeling of cosy warmth, comfort, and togetherness; drinking tea with friends by candlelight is hygge, but it cannot be defined.

Each is a conceptual tool that concentrates its speakers’ attention on a particular domain of experience. It does not mean that the experience cannot be had without the word – but with the word, the experience is named, shared, passed from generation to generation, and a culture forms around it.

Every Language Is a Window – But You Need a Single Home to Think In

The examples shown so far establish a truth: every language is a window viewing the universe from a different angle. Russian divides the colour space differently. Kuuk Thaayorre directs time differently. Turkish makes the source of knowledge visible. Chinese engraves history into its characters. Japanese compresses the beauty of transience into a single word. Portuguese can even feel longing for the future.

These riches are real. To fully enter those worlds, you need to know those languages – because translation transfers the structure of the word but cannot transfer the experiential network beneath it. You can explain what saudade is in English, but being able to feel saudade in Portuguese is a different experience. You can read the definition of komorebi, but when a Japanese speaker hears the word and their mind instantly summons that image, that play of light, that sense of transience – that is a lived language experience.

Does this mean you need to be educated in that language to think?

No. And it is precisely at this point that this essay’s thesis takes its clearest form.

Every language’s window shows that language’s world. But thinking – comprehending, analysing, synthesising, producing – these are done not by looking through the window but inside the house. And that house is the mother tongue.

Knowing Japanese to understand wabi-sabi is a fine thing. But to think deeply in any field – engineering, law, medicine, philosophy – knowing English is not necessary. Access to knowledge is necessary – and that access is ceasing to depend on language (as Section VII will show). But thinking itself – the inner space where inner speech occurs, where concepts form, where arguments are constructed – is the mother tongue.

The richness of languages is not an argument in favour of English-medium education. On the contrary: every language’s unique conceptual world is proof of why thinking in the mother tongue is indispensable. Turkish’s evidential markers enable a person who thinks in Turkish to automatically interrogate the source of their knowledge – to lose this cognitive advantage by switching to English-medium education is to demolish the house while opening a window.

VII. Artificial Intelligence and the New Equation

Three Channels, Three Limits

Knowledge reaches people through three traditional channels. Each has a structural limit.

Book translation is static. A translator translates a text once – and from that moment on, the text is the same for everyone. It does not know who the reader is, what they know and do not know, through what conceptual framework they think. Moreover, the translator is bounded by the same limits this essay has described: they can transfer the first and second layers of the source text but mostly cannot transfer the third and fourth – the cultural nuance, the philosophical background, what the author implies but does not say. You translate the structure of a concept, but the cultural sediment beneath it does not clear customs. And the translator translates “for everyone” – presenting the same standard, depth, and framework of information flow to all. A lawyer and an engineer receive the same translation; because there is no other way.

The teacher is live but limited. If they are lecturing in English – however proficient they may be – they are spending part of their cognitive capacity on language processing. The mechanism this essay has described applies to the teacher too: the capacity for improvisation, conceptual flexibility, and emotional depth of a teacher lecturing in a foreign language is reduced. The “panic” documented by Zheng and Choi (2024) – the first year of teaching in a foreign language – is the tip of the iceberg. And in a class of 30, the teacher faces the same structural problem as the translator: they give the same lesson to everyone. Even in one-to-one tutoring – even focused on a single student – they cannot truly see that student’s conceptual world. They guess, observe, try to adapt – but human memory, attention, and time are finite.

Benjamin Bloom’s (1984) “2 sigma problem” measures this limit: in Bloom’s original measurement, a student receiving individual tutoring outperformed a classroom student by 2 standard deviations. Later studies found the effect more moderate (roughly 0.8 standard deviations for human one-to-one tutoring; VanLehn, 2011), but the superiority of one-to-one tutoring is beyond dispute. The problem is that a personal tutor cannot be provided for everyone. For forty years, this problem remained unsolved.

Interactive AI is categorically different from these two channels. The difference is not one of scale but of structure.

Not Translating, but Carrying

The writing process of this very essay is a living proof of its own thesis. Most of the academic sources cited here are in English. The tool used to access, analyse, and integrate them into the argument: artificial intelligence. This was not possible five years ago.

But to describe what AI does here as “translation” is to remain at the first layer.

A translator converts “death” to the target language equivalent – transferring the word. Interactive AI does something different. Taking into account what the user is looking for, in what context they are asking, through what conceptual framework they think, it places the knowledge into the user’s world. It carries not the word’s counterpart but the meaning itself – adapted to the user’s own conceptual network, their own metaphor system, their own experiential layers. It presents the Keysar experiment to a lawyer through the framework of legal reasoning; it explains the same experiment to an engineer through cognitive load theory; it shows a parent the practical consequences. The same knowledge enters different worlds through different doors.

In this process, the knowledge does not change. The Keysar experiment is the same experiment; Bälter’s data is the same data. What changes is not the knowledge itself but the conditions of encountering knowledge. And there is a critical distinction here: the knowledge is not personalised; the process of understanding is personalised. The translator presents the same text to everyone – the same standard. Interactive AI works in direct, context-based communication with each user and can present at different depths according to that user’s questions, level of knowledge, and mode of thinking.

Bloom’s 2 sigma problem becomes relevant again precisely at this point. The difference created by a personal tutor was real but unscalable. Interactive AI removes this structural limit: one-to-one interaction with each user, unlimited patience, continuous context tracking. For a teacher to achieve personalisation at this scale, they would need to focus on a single student, know them for years, and devote their entire cognitive capacity to that one person – and even then, the limits of human capacity apply.

Limits and Honest Distinctions

This does not mean AI “solves everything.” It has two serious limits.

First: AI does not understand the knowledge. The mind that understands, thinks, and connects concepts is still the human mind. What AI does is remove the obstacle before the process of understanding – the language barrier – and improve the conditions of understanding. Elzerman (2025) warns of the “cognitive offloading” risk: when AI takes over thinking, people rely less on their own analytical skills. Derakhshan and Taghizadeh (2025) showed that AI can weaken higher-order thinking skills. AI can carry knowledge, but it cannot carry thinking – the thinking is still done by a human, and that human must maintain their thinking capacity.

Second: the layer at which AI carries knowledge is still debatable. Chua and colleagues’ (2024) study examining multilingual large language models is cautionary: these models show surface-level crosslingual competence but face serious barriers in deep knowledge transfer. Performance drops markedly in languages other than English. Ray’s (2025) finding is more intriguing: when the same question is posed to AI in different languages, it is answered with different framings and emphases. The linguistic relativity hypothesis measurably affects even AI-generated content – machines, too, encounter the core distinction of this essay.

Stanford HAI’s (2025) report adds a pragmatic caveat: AI works well for the 1.52 billion English-speaking people but underperforms for low-resource languages. Data inequality sustains the language barrier even in the AI revolution.

The first and second layers – word decoding and inferential understanding – are areas where AI is strong. The third layer – cultural nuance – is developing but has gaps. The fourth layer – deep understanding, accessing philosophical assumptions – how far AI can reach this layer remains an open question. And that open question stands at the centre of the next study.

Paradigm Shift: Yesterday’s Validity, Today’s Question

The previous generation’s argument was right. In the 1970s, 1980s, and 1990s, not knowing a foreign language truly meant disconnection from knowledge. The only way to access knowledge was to read the book, article, or newspaper in that language. Translation was slow, expensive, limited. Academic resources in Turkish were scarce. The channels of that era – book translation and the teacher – truly were the only routes. In that era, saying “you cannot access knowledge without knowing a foreign language” rested on a reality.

But no one is asking this question: do the conditions of that era still hold?

Two things have changed; one thing has not.

What changed first: access to knowledge. Frey and Llanos-Paredes’ (2025) Oxford study, examining 695 US local labour markets, documents that the spread of machine translation has effectively reduced demand for foreign language skills. Shadiev and Huang’s (2020) experiment confirms this in an educational context: when foreign-language lectures were supplemented with mother-tongue transcription, student performance significantly improved, especially among students with lower language proficiency. Cognitive load decreases because the brain can allocate its energy to thinking instead of translation.

What changed second: the mode of encountering knowledge. The old channels – book translation and the teacher – presented knowledge in a standard form: the same text for everyone, the same lesson for everyone. Interactive AI can present knowledge according to the user’s world. This is a change beyond access – and the dimensions of this change are not yet fully understood.

What has not changed: thinking’s dependence on language. People still think, comprehend, and formulate in their mother tongue. All the data shown throughout this essay – Keysar’s foreign language effect, Dewaele’s inner speech study, Bälter’s randomised experiment – all point to this unchanging reality. And the cross-linguistic differences shown in Section VI – Turkish’s evidential markers, Russian’s shaping of colour perception, Kuuk Thaayorre’s directing of time – all reinforce the same point: language is not merely a communication tool; it is thinking infrastructure. Changing that infrastructure is like moving house – not changing windows.

If, then, access to knowledge no longer requires English, and the mode of encountering knowledge has moved beyond standard translation, what justification remains for moving thinking into English – that is, for English-medium education?

Stated more sharply: if access to knowledge and the conditions for comprehending it can be provided through a tool far more comprehensive than English, the rationale for making English the language of instruction must be re-examined. And if AI can place the world’s libraries before everyone in their mother tongue, adapted to their world, the assumption that being global requires going through English may itself be wrong.

Two Models, Two Architectures

The emerging picture brings two different knowledge-transfer architectures face to face.

English-medium instruction model: Moves both access to knowledge and thinking into the foreign language – both at once. Divides brain energy between translation and thinking, and as Bälter showed, this division can cost the thinking side up to 41%. Knowledge is presented in a standard form, at a standard depth, the same way to everyone. And the energy allocation decision is not left to the student – the system makes this decision on their behalf.

AI-supported mother-tongue model: Separates the two – AI solves the access problem, thinking stays in the mother tongue. All cognitive energy can be devoted to comprehension, analysis, and synthesis – because the machine bears the translation load. And this machine, unlike a translator or teacher, adapts to the user’s world. If cognitive energy is a finite resource – and it is – where you spend it is decisive. This model directs that finite resource to thinking; the other directs it to translation.

The fact that a researcher who does not know English can access English-language academic sources through AI and conduct their analysis in their mother tongue – this very essay is a living application of this model. The fact that this phenomenon has not been systematically investigated – no documented study specific to Turkey was found – is a gap. The gap itself is a datum: no one has asked.

That no one has asked may be the real subject of this essay. And the answer to this question – whether knowledge can truly be liberated from the language in which it is imprisoned, and when liberated, whether it can be carried all the way to the fourth layer or remains at the first two – is the subject of the next study.

VIII. The True Place of English

The World’s Language

This essay is not a manifesto against English. We cannot ignore the fact that English is the world’s language of global communication – the language of science, of business, of diplomacy, of the internet. The vast majority of academic publications are in English. International conferences are in English. Software documentation is in English. This is a fact.

But as Section VI showed, English, like all other languages, is a language with its own window. The “I see” = “I understand” metaphor system is a way of thinking specific to English, one that constructs knowledge through sight. The absence of a word for Schadenfreude in English does not mean English speakers do not experience that feeling – but it means they have not named it, have not culturally recognised it. English, too, cannot open all windows simultaneously.

This fact does not prevent us from asking a different question: what do we use English for?

English as a language of communication: Necessary, beyond debate. As a meeting point for two people who speak different mother tongues, English is indispensable to the world. No one disputes this.

English as a language of knowledge access: Increasingly less necessary. AI is rapidly removing this barrier. Five years ago, a Turkish researcher needed either to know English or to find a translator to access an English-language paper. Today they can read it within minutes.

English as a language of thinking: This is where we must stop. This essay’s entire argument converges at this point. The rate of thinking in English, according to all the empirical data, is extremely low. Even at C1–C2 levels, inner speech is predominantly in the mother tongue, emotional processing in the mother tongue, mental imagery in the mother tongue, automatic responses in the mother tongue. “Thinking” in a second language means, for most people, thinking in the mother tongue and then translating into the second language – thinking itself does not leave the mother tongue.

Communication ≠ Understanding ≠ Thinking

These three concepts are constantly conflated:

I can communicate = I can make my words understood. This requires BICS-level proficiency.

I understand = I can grasp what is meant. This requires pragmatic competence and cultural depth – acquired not through education but through living.

I can think = I can construct arguments, develop counter-positions, produce creative solutions in that language. This requires near-native-level proficiency – beyond the reach of most people.

The problem with Turkey’s English-medium education debate is that all three levels are called “knowing English.” A graduate of an English preparatory class says “I know English.” A person with a C1 certificate says “I know English.” And a literature professor who can interpret Shakespeare in English also says “I know English.” All three use the same phrase, but their “knowing” is completely different.

IX. Objections to the Thesis

Objection 1: “Those educated in English are still more successful.”

The observation is correct. But the causation is wrong. A student who enters Boğaziçi University’s Economics programme is one of the top 657 people in Turkey. A student entering Istanbul University’s Turkish-medium Economics programme is ranked 115,803rd. The gap is 176-fold. Comparing these two graduates and saying “English-medium education makes the difference” is the same as comparing an Olympic athlete with a neighbourhood league player and saying “the training programme makes the difference.” The difference is made by who walks through the door in the first place.

Civan and Coşkun (2016) showed this directly: when entry scores are controlled, the advantage of English-medium education disappears or reverses.

Objection 2: “Bilingualism confers cognitive advantages.”

Nichols and colleagues’ (2020) population study of 11,000 individuals: no evidence was found that bilingualism provides a general cognitive advantage. Masullo, Dentella, and Leivada’s (2024) systematic review: 73% of bilingual (dis)advantaged cognitive effects stem from sociolinguistic factors. The true cognitive effect largely vanishes when socioeconomic status is controlled.

Objection 3: “The problem is with the quality of education, not the language.”

This is the strongest counter-argument. I largely agree. If English-medium instruction were properly implemented, the outcomes might differ. But if the system is not working for the majority – and the data shows this – then the “implementation problem” defence becomes a way of avoiding questioning the system itself.

Objection 4: “English-medium education provides a career advantage.”

It does. But how much of that advantage stems from genuine competence and how much from the “label”? A Boğaziçi diploma is a career advantage – but does that advantage come from English-medium instruction, or from Boğaziçi’s brand and its high entry threshold? Selection effects are at play here too.

Objection 5: “AI has not yet fully removed the language barrier.”

True. Deep knowledge transfer barriers exist in particular (Chua et al., 2024). Literary translation, nuanced legal text, cultural context – AI still falls short in these areas. But this does not mean AI is “useless.” A revolutionary change in knowledge access is under way and accelerating.

Objection 6: “If every language has its own window, shouldn’t we learn English’s window too?”

This is a strong objection, the natural extension of Section VI. The answer: yes, looking through English’s window is a fine thing – just as looking through the windows of Japanese, Chinese, or Portuguese is. But none of these windows is necessary for thinking. Learning English is an enrichment – being educated in English means moving one’s thinking infrastructure into that window. There is a difference between looking through a window and moving the house.

And if the aim is to multiply windows, why only English? Why not Japanese, Chinese, German? The answer is clear: the motivation is not multiplying windows but the label of “global citizen.” The system is selling not English’s window value but its brand value.

The Correct Formulation

This essay is not saying “English-medium education is bad.” It is not saying “do not learn English.” The person who learns a language out of love for it, to live in it, to enter that language’s world is an exception – and no one objects to this exception. That person acquires the language not as a tool but as a mode of thought and an art; they are not the addressee of this essay. This essay’s addressee is the ninety-nine percent – the great majority forced to learn the language as a tool, unaware of what that tool gains and loses them.

What is being said is this:

Turkey’s English-medium education system does not deliver on its promise. The promise: “a global citizen who can think in English.” What is delivered: individuals who can decode words but cannot reach depth, lost in the illusion of competence, their thinking potential pruned – and institutions that climb a few rungs in university rankings.

X. Not Knowing a Language Is Not a Handicap

The person who knows they do not understand develops compensatory mechanisms – they consult a translation, ask an expert, read the context, apply logic, verify. The person who believes they understand develops nothing.

The greatest cost of shallowness mistaken for mastery is not the loss of knowledge – it is the invisibility of knowledge loss. And if cognitive energy is a finite resource, where you spend it is decisive. Energy spent pursuing an unreachable proficiency is energy not spent on thinking – and that energy does not come back.

We need to use the word “understand” more carefully. Decoding a word is not understanding. Translating a sentence is not understanding. Following a conversation is not understanding. Understanding is seeing the world behind a thought – and this is a skill that is difficult even in one’s own language.

The rate of thinking in a second language, according to all the empirical data, is extremely low. Even at C1–C2 levels, people predominantly think, feel, dream, and calculate in their mother tongue. English-medium education does not change this reality – it merely hides it.

Perhaps the problem is not with English-medium education. Perhaps the problem is that we do not understand the concept of “understanding.”

This essay has drawn the map of that problem. The next will ask what it means to liberate knowledge from the language in which it is imprisoned – whether knowledge that a translator presents identically to everyone can truly be carried into an individual’s own conceptual world through interactive AI. And whether that carrying operation stops at the first layer or can reach the fourth – that is what it will test.

Questions are more important than answers.

Halit Cengiz Uzuner · Independent Researcher · halitcengizuzuner.com