Two Kinds of Knowing: Language, Technology, and What Gets Left Behind

There is a reasonable argument that language learning is becoming unnecessary.

Apple's AirPods now translate speech in real time. Google Translate handles written text instantly in over a hundred languages. Wearable devices with integrated translation are already in development, and the direction is clear: the gap between not knowing a language and functioning within it is closing fast.

This is worth taking seriously. If you land in a country whose language you do not speak, real-time translation is genuinely useful. It lets you ask for directions, negotiate a price, follow a conversation. It removes barriers that once required months of study. For short-term contact across linguistic borders, the technology works, and saying otherwise would be dishonest.

But something else is happening at the same time, and it concerns the nature of knowledge itself.

What Language Learning Does to the Brain

When a person learns a language, something changes in the brain. This is not a metaphor. Structural neuroimaging studies have confirmed that sustained engagement with a second language physically alters brain tissue in regions associated with language control, executive function, and cognitive switching. A 2023 study published in Scientific Reports (Korenar et al.) documented these changes using MRI data from bilinguals and found that the brain follows non-linear adaptation trajectories similar to those seen in other cognitively demanding skills. The brain expands in certain regions and then renormalizes, a pattern consistent with experience-based neuroplasticity more broadly.

The region most associated with memory formation and contextual learning, the hippocampus, is among those that respond to bilingual experience. A 2025 study in eNeuro measured gray matter volume in young adults and found an inverted U-shape relationship between second language engagement and left hippocampal volume: tissue expands with increasing engagement and then renormalizes once the skill is acquired. This non-linear pattern is consistent with how the brain adapts to other demanding skills. It is not a simple equation of more practice equals more tissue. It is evidence of the brain actively reorganizing itself in response to sustained cognitive demand. These are not minor adjustments. The brain of someone who has learned and actively used a second language is built differently from what it was before.

A translation earpiece does not produce these changes. It mediates communication without requiring the neurological work that produces them. The person who speaks through a device remains cognitively in one language while the device handles the other. The brain is not being restructured. It is being bypassed.

Language Is Not a Code for Content

The question is whether that matters, and for what.

If the goal is transactional communication, it probably does not matter much. Two people who do not share a language but need to exchange information can do so through a device efficiently. The information moves and the task is completed.

The goal shifts when communication involves something other than information exchange. A sentence carries more than its literal content. It carries the speaker's relationship to what they are saying, the social distance between the people talking, the cultural assumptions behind why certain things are said and others left unsaid.

Automatic translation moves the words. It does not move any of that. What arrives on the other side is a grammatically functional sentence stripped of the layer that gave it weight.

This is where research on linguistic relativity becomes relevant. The strong version of the Sapir-Whorf hypothesis, that language determines thought absolutely, has been largely rejected. The weak version, that language influences perception and cognitive processing in measurable ways, has accumulated substantial empirical support since the 1990s. Studies have shown that speakers of different languages attend to spatial relationships differently, perceive color categories differently, and mark causal chains differently, depending on what their grammar requires them to encode.

When you move between languages, you are not swapping labels on the same underlying concepts. You are shifting between systems that organize experience differently. The grammar is not packaging for thought. It is part of how thought takes shape.

What Turkish Grammar Encodes

Turkish makes this visible in specific, traceable ways.

Consider the evidential suffix system. When a Turkish speaker says something happened, the grammar itself encodes the source of that knowledge. Gördüm means I saw it myself. Görmüş means I heard it, or I am inferring it from evidence. English does not grammaticalize this distinction at all. You can add a qualifier in English, "I heard that..." or "apparently," but it is optional. In Turkish, it is not. Every time a speaker reports something, the language asks: how do you know this?

A learner who reaches the point where this distinction feels automatic has not acquired a vocabulary item. They have internalized a different relationship to knowledge and testimony. The grammar has trained them to track the source of what they know.

Or consider how Turkish handles the sequence of thought within a sentence. In English the verb comes early. The predicate is established close to the subject, and then qualified. In Turkish, the main verb comes last. Everything else builds toward it.

The entire weight of the sentence lands at the end. A learner who becomes fluent in Turkish builds a cognitive habit through thousands of sentences: to hold things in suspension, to defer resolution, to carry context further before releasing it.

There is also the matter of what Turkish does not encode that other languages do. Turkish has no grammatical gender. Objects, animals, professions, strangers encountered in a story: they are all referred to with the same neutral pronoun, o.

Research on grammatical gender in other languages has found that speakers do attribute gender-related qualities to inanimate objects along the lines of their grammatical gender. Turkish does not activate this. A learner coming from a gendered language and acquiring Turkish stops applying a categorization that was previously automatic.

A translation device can output the equivalent sentence in English. The person wearing it never passes through any of this. They never feel the difference between gördüm and görmüş as a distinction the language itself demands. They never arrive at the end of a long Turkish sentence and feel the resolution that comes with the final verb. They receive the output without doing the cognitive work that produces it, and the cognitive work is precisely what changes how a person thinks.

But there is a layer beyond cognition that matters here too.

Language is a cognitive system and also a social practice. When you learn another language well enough to actually live inside it, you are acquiring grammar and something beyond it. You are learning how to participate in a different set of social relationships. You are learning how to position yourself: how to sound respectful without being distant, how to express uncertainty without losing credibility, how to be warm in the specific way that language expects warmth to sound.

Turkish has a register system built into its pronouns and verb forms. The difference between sen and siz, between addressing someone as an equal or as a figure of deference, is not a formality you can look up and apply. It is something you absorb through thousands of real interactions, through getting it wrong, noticing the slight shift in someone's expression, and adjusting. A device does not teach you that. And more to the point, a device cannot make you feel the social weight of choosing correctly or the discomfort of choosing wrong. That discomfort is part of the learning.

Learning Turkish means becoming, at least partially, a different kind of social participant. You develop a voice in the language, not just a translation of your existing one. That voice is built from grammatical choices, from cultural exposure, from the slow internalization of what Turkish speakers expect from each other in a conversation. It cannot be borrowed from a device. It has to be built.

The Effort Problem and Why Ease Is Not Neutral

This is where I want to be precise about something, because it is easy to misstate.

The argument is not that technology is bad or that translation tools should be avoided.

The argument is that using a tool and internalizing a skill are different cognitive events, and the difference between them is not just a matter of degree.

Research on cognitive offloading, the practice of delegating memory and processing to external tools, shows a consistent pattern. Offloading improves immediate task performance and frees working memory for other things. But it also reduces the depth of encoding for the offloaded content. When you let a device remember something, you tend not to remember it yourself. Studies in this area have found that heavier offloading improved task completion while reducing subsequent memory accuracy for the same material. The device performed better. The person retained less.

Applied to language: when the tool produces the sentence, the learner does not build the internal structure that produces the sentence. Recognition increases while production stays low, and the knowledge remains outside the person.

Robert Bjork's research on desirable difficulties, examined again in a 2024 comparative analysis published in Quarterly Journal of Experimental Psychology (Pyke et al.), shows that conditions that feel harder during learning tend to produce stronger long-term retention. Retrieval practice, spaced repetition, the effort of constructing a sentence without support: these are cognitively costly in the moment, and they produce more durable encoding because of that cost. Easy access removes the friction that builds the structure. A language acquired with effort is held differently than a language processed through a device.

A 2025 study on AI tool usage and critical thinking, published in MDPI's Societies (Gerlich), found a significant negative correlation between frequent AI tool reliance and critical thinking scores, mediated by cognitive offloading. The tool-dependent group performed tasks. The less tool-dependent group understood them. That gap is worth paying attention to.

The Attention Economy and the Intolerance for Slow

There is a second force working alongside technology that rarely gets named directly in discussions about learning: the effect of short-form content on the capacity to tolerate difficulty.

When every piece of content is optimized to deliver a reward within seconds, a different standard becomes the baseline. The brain learns to expect resolution quickly.

Anything that does not deliver that resolution starts to feel like failure, or at minimum, like inefficiency. I see this regularly in my own work. When I produce content for social media, if the information does not land within ten seconds, the comments follow: too long, why so much talking, get to the point. The discomfort is not with the content. It is with the duration. This is the dopaminergic logic that platforms like TikTok and Instagram Reels operate on. The content is structured to produce anticipation and satisfaction in very short cycles, and sustained use trains the nervous system to expect that rhythm.

Language learning runs on a completely different rhythm. It is slow by design. A suffix rule understood today will not feel natural for weeks. A sentence that requires real-time construction under conversational pressure will fall apart before it gets easier.

The discomfort is not a sign that something is wrong. It is how the encoding happens.

But a person whose attention has been trained to flee discomfort within three seconds is not well-positioned to sit with that process.

The environment has been designed to reward impatience, and the results are predictable. The appetite for fast, immediately satisfying knowledge has always existed in human beings. What has changed is an entire industry built to feed it at scale, and the cognitive cost of that feeding.

The learner who gives up on Turkish after two months because it feels hard, and installs a translation app instead is not making an irrational choice within the logic of the environment they inhabit. The environment has been designed to make slow effort feel like the wrong option. What gets lost is not just the language. It is the capacity for the kind of attention that slow learning requires. And that capacity, once eroded, affects everything else too.

When Language Becomes Infrastructure

Per Urlaub and Eva Dessein, researchers at MIT who published a review of machine translation in foreign language education in Frontiers in Artificial Intelligence (2022), raised a concern that has stayed with me. They argued that if deployed as a primary instructional technology, machine translation tools may produce what they called "reductionist perceptions of language" among students and teachers alike.

Language begins to look like a code for transferring content between containers. Grammar becomes invisible infrastructure. Cultural weight disappears.

I find this framing precise because it describes something I see happening outside the classroom too. When language is reduced to its function as a transmission channel, a natural conclusion follows: the channel should be as efficient as possible, and if a machine runs it better than a human, then human effort is wasted. The question of whether to learn stops being a question about what kind of person you want to become and turns into a question of efficiency.

This is why I do not teach Turkish as a tool. I teach it as a cultural system. The language carries history, perception, social structure and ways of knowing that have no equivalent in English and no proxy in translation. When a student learns the word gurbet, they are not acquiring a synonym for "homesickness." They are entering a concept that Turkish culture has refined over centuries: the particular weight of being far from your place of origin, the sense of displacement that does not disappear even when the distance is geographical rather than emotional. Gurbet is a relationship to belonging. No device translates that. It is only accessible through the language itself, and even then, only gradually, through use.

Or consider hüzün, which Orhan Pamuk described in Istanbul as a collective melancholy, a mood embedded in the city's relationship to its own history. The word is often translated as "sadness" or "melancholy." It is neither. It is something closer to a shared aesthetic of loss, a feeling that is communal rather than personal, beautiful rather than simply painful. A learner who encounters hüzün in text, asks a translation device for its meaning, and receives "melancholy" has not learned the word. They have been given a placeholder that makes the word invisible.

The grammar is not packaging. Vocabulary is not a word list. The cultural layer is not background information that can be separated from the linguistic content and added later. They are one thing, and encountering that one thing is what learning a language actually is. When that process is bypassed, what gets bypassed is not inconvenience.

It is the substance.

The Divide That Is Forming

This brings me to what I think is the real divide forming now, and it is not the one usually described in discussions about AI and jobs or AI and creativity.

The divide is between two relationships to knowledge. In one, knowledge is a resource: accessible, usable, replaceable. You reach for it when you need it and put it down when you do not. Technology makes this kind of knowing more efficient and cheaper. Translation earbuds belong here. So does looking up a formula rather than memorizing it, or asking a chatbot to draft a text rather than writing it.

In the other relationship, knowledge is a process that the person passes through and does not emerge from unchanged. Learning a language belongs here, but so does learning to play an instrument, or developing a craft, or reading slowly in a second language until the syntax begins to feel like something other than a puzzle. These are slow accumulations that alter the person doing them.

The cost of the first kind of knowing is falling. The cost of the second is not, because it was never about access to information. It was always about the time spent doing the cognitive work that transforms a person rather than just equipping them.

What technology is doing is not replacing this second kind of knowing. It is making the first kind so available and so adequate for so many purposes that the motivation to pursue the second declines. Why learn Turkish if a device translates it? The answer cannot be "because you might lose your phone." It has to be something more honest. The process of learning a language or learning anything changes how you think, what you notice, and what categories you use to parse experience. The language, learned slowly and with effort, passes through you and leaves you different on the other side.

The brain is not a hard drive that records and holds still. It breathes. It reorganizes itself around what you ask of it. A person who learns a language is not the same person who began.

What the Shift Looks Like in Practice

I have watched students arrive at a point in Turkish where something shifts. The language stops being a set of rules they apply and starts being a medium they move through. The suffixes stop feeling like additions and start feeling like parts of a thought. This transition takes time, and it requires the kind of effort that does not feel productive in the moment but produces something that cannot be replicated by retrieval. It produces a person who thinks differently because they learned to.

That is the kind of knowing that technology cannot make easier, because the difficulty is the mechanism. Remove the effort and you remove the transformation.

The technology will keep improving. Real-time translation will become more accurate, more subtle, and more socially embedded. It will handle more of the surface of communication and handle it well. For many people and many purposes, that will be enough.

The question worth sitting with is what it means to be a person who chose to go further anyway.

---

Frequently Asked Questions (FAQ)

Q: If translation technology keeps improving, is there any practical reason left to learn a language?

A: For purely transactional purposes, the practical case is genuinely weaker than it was twenty years ago. But the argument for learning a language was never only practical. The process changes how you think, how you track the source of what you know, how you position yourself in social situations. Those changes happen through the learning, not through the output. A device that produces correct sentences does not produce those changes in you.

Q: Isn't this just an argument that difficulty is good for you?

A: Not exactly. The argument is more specific: certain kinds of knowledge are only produced by the effort of acquiring them. The difficulty is not a character-building exercise added on top of the learning. It is the mechanism by which the encoding happens. Research on desirable difficulties in cognitive psychology makes this precise: retrieval under effort produces stronger retention than retrieval with support.

With language specifically, the effort of constructing a sentence without a device is what builds the internal structure that makes the next sentence possible.

Q: Can someone use translation tools and still learn a language properly?

A: Yes, if the tools come after the attempt, not before it. The difference is significant. Using a device to check a sentence you already tried to build is different from using it to avoid building the sentence at all. The first keeps the cognitive work on your side. The second moves it outside you.

Q: Does this mean short-form content and social media are bad for language learners?

A: They are useful for exposure and for staying in contact with the language. The problem is specifically about attention: what sustained platform use trains the nervous system to expect. A learner who watches Turkish content on social media is getting genuine input. A learner whose capacity to sit with difficulty has eroded through years of fast-reward content will find that input harder to convert into production. These are different issues, but they interact.

Q: You mention that Turkish has no grammatical gender. Does that actually change how speakers think?

A: The research on grammatical gender suggests that it does influence how speakers categorize objects and assign qualities to them, though the effect is contextual rather than absolute. Turkish speakers who acquire a gendered language later in life report the category as strange at first, not because they do not understand it but because it requires tracking a distinction the brain did not previously need to maintain. Learning in the other direction, a gendered-language speaker acquiring Turkish, involves something similar: releasing a categorization that was previously automatic. That release is a cognitive event, not just a grammatical one.

Q: Is there a version of this argument that applies to things other than language?

A: Yes. The divide between knowledge as a resource and knowledge as a process that changes you applies anywhere that effort is the mechanism of transformation. Music is the clearest parallel. A person who listens to a great deal of music has very different hearing from a person who learned to play. Both are engaging with the same art form. One of them is being restructured by it.

References

Korenar, M., Treffers-Daller, J. & Pliatsikas, C. (2023). Dynamic effects of bilingualism on brain structure map onto general principles of experience-based neuroplasticity. Scientific Reports, 13, 3428. https://doi.org/10.1038/s41598-023-30326-3

Gallo, F. et al. (2025). Experience-Dependent Neuroplasticity in the Hippocampus of Bilingual Young Adults. eNeuro, 12(6), ENEURO.0128-25.2025. https://doi.org/10.1523/ENEURO.0128-25.2025

Pyke, W., Lunau, J. & Javadi, A.-H. (2024). Does difficulty moderate learning? A comparative analysis of the desirable difficulties framework and cognitive load theory. Quarterly Journal of Experimental Psychology, 78(10), 2181–2195. https://doi.org/10.1177/17470218241308143

Urlaub, P. & Dessein, E. (2022). Machine translation and foreign language education. Frontiers in Artificial Intelligence, 5, 936111. https://doi.org/10.3389/frai.2022.936111

Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), 6. https://doi.org/10.3390/soc15010006