Language Wasn't Designed
Pierre Teo · Updated · Language learning
For a long time, I assumed language was a top-down, rules-based system. I assumed that somewhere behind English, Thai, or Japanese, there was a coherent master specification, and fluent speakers had simply internalized the rulebook. Learning a second language, then, seemed like a matter of studying the rules, memorizing the vocabulary, and executing them.
That false impression did not come from how speech actually works: it came from traditional classrooms, grammar drills, and standardized exams. Because we don't remember learning our first language, school was our only conscious memory of how language is acquired. We were taught to treat language like an academic subject to analyze.
In computing, programming languages really are engineered from formal specifications. A standards committee defines the syntax, writes the compiler, and rejects any ambiguity. But code is designed for computers. Natural language evolved for humans.
Nobody designed English, Thai, or Japanese. Nobody sat down and wrote the grammar rules first. Natural language grew the way an ancient city grows: unplanned, sprawling, layered over centuries, and shaped by millions of ordinary speakers adapting it to real life.
The word that does twenty jobs#
Consider a word as mundane as "get":
- I get it. (understand)
- I got home. (arrive)
- I'm getting tired. (become)
- Get out. (leave)
- Get along. (cooperate)
- Get over it. (recover)
- She gets on my nerves. (irritate)
None of these meanings share a clean dictionary definition. "Get" is doing completely different conceptual work in each sentence, yet native speakers process the correct meaning in milliseconds without the slightest pause.
If an engineer tried to design a programming language with that level of overloading, it would be rejected immediately as unmaintainable chaos. Yet in human speech, it works flawlessly.
A living geological record#
Every natural language is a history of cultural collisions fossilized in vocabulary.
English, for instance, accumulated over fifteen centuries of invasions, migrations, and social shifts. When Germanic Anglo-Saxons farmed the land, they called the animal in the pasture a cow, sheep, or swine. When the French-speaking Normans conquered England in 1066, the meat served at aristocratic banquets became boeuf (beef), mouton (mutton), and porc (pork). Both vocabularies survived in parallel because two social classes were speaking side by side.
Centuries later, the Great Vowel Shift fundamentally altered English pronunciation while the invention of the printing press froze spellings in place. That is why knight retains its silent k, and why through, though, thought, tough, and thorough look almost identical while sounding completely different.
Spelling and grammar inconsistencies aren't bugs; they are geological strata.
Even artificial attempts to engineer a clean, perfectly regular language prove the point. In 1887, L.L. Zamenhof created Esperanto: completely regular conjugations, phonetic spelling, and zero exceptions. But something fascinating happens when children grow up speaking Esperanto natively at home: they begin introducing their own irregularities. They shorten common phrases, bend grammatical rules for rhythm or emphasis, and let context do the work.
Strict logical consistency does not survive contact with real human lives.
Why the messiness actually works#
Language is not optimized for mathematical elegance. It is optimized for something far harder: real-time communication between imperfect human brains across noisy rooms, limited attention spans, and unpredictable situations.
What looks like messy design is actually an extraordinary set of engineering tradeoffs:
- Ambiguity is compression. If every nuanced concept required a unique, dedicated word, human vocabulary would require millions of entries. Instead, language relies on the brain's incredible context engine. A single flexible word like "run" (run a marathon, run a business, run out of battery) does dozens of jobs because surrounding context instantly disambiguates the meaning.
- Frequency wears paths smooth. The most common words in almost every language are the most irregular (be/was, go/went, have/had). High frequency either preserves archaic forms or wears syllables down into faster, easier vocal shapes. The phrases used most are the ones that drift farthest from the textbook rules.
- Redundancy ensures error tolerance. An engineered network protocol might strive to transmit each bit of data exactly once. Human speech does the opposite. Tone of voice, facial expressions, grammatical suffixes, and sentence order all partially encode the same message. If ambient noise drowns out half a sentence, or if someone mispronounces a word, the listener still understands the intent.
The statistical fabric#
Because language was never designed top-down, we cannot master it by studying top-down rules.
Textbook grammar rules are not the source code of language. They are after-the-fact descriptions written by linguists attempting to document habits that already existed for centuries. Humans spoke grammatically for millennia before the first grammar book was ever published.
The true structure of language is statistical:
- Certain sounds consistently follow other sounds.
- Certain words naturally pair together (collocations like "fast food" vs "quick shower", "make a decision" vs "take a chance").
- Certain phrasings imply meaning far beyond their literal definitions ("Can you pass the salt?" is an action request, not an inquiry into arm strength).
Native speakers don't know the rules because they memorized a spec. They know the patterns because their brains spent thousands of hours computing the statistical likelihood of what belongs where.
What this means for learning#
Once we accept that language is an evolved ecosystem rather than an engineered blueprint, our approach to learning completely changes.
We have to stop treating a target language like programming syntax to compile in our heads. We cannot execute grammar rules in real time during a live conversation.
Instead, we have to do what the brain was built to do: feed it massive amounts of real, contextualized patterns. Read and listen to speech we can understand, observe how native speakers express real ideas, and let the brain's subconscious pattern engine extract the structure naturally.
There is no master blueprint to memorize. There are only patterns that survived because they worked.