Strong tea, powerful engine: the hidden rules of which words go together

In English we drink strong tea, never powerful tea. But we buy a powerful engine, not a strong engine. Both alternatives are grammatical; a listener would understand you perfectly either way. Yet one of each pair sounds like a native speaker and the other sounds like a translation. Nothing in a dictionary definition of strong or powerful tells you which is which. The rule lives somewhere else entirely — in the habits words have about the company they keep.

This is collocation: the tendency of certain words to turn up next to certain other words far more often than chance would predict. We make a decision but do the dishes. Rain is heavy, not strong. A crime is committed. These are not logical facts about tea or rain or crime. They are conventions — the quiet, unwritten agreements a language has made with itself about which words are friends. And they are a far larger part of fluency than most learners ever suspect.

You shall know a word by the company it keeps

The idea has a famous one-line summary. In 1957 the linguist J. R. Firth wrote that "you shall know a word by the company it keeps," and with that sentence gave the study of collocation its motto. His point was practical as well as poetic: part of what a word means is simply the set of other words it habitually appears beside. You cannot fully know rancid without knowing it goes with butter and not with milk; you cannot fully know heavy until you know it can weigh down rain, traffic, and a smoker's habit, but not tea.

For a long time this looked like a charming footnote to the real business of grammar and vocabulary. Then people started counting, and the footnote turned out to be most of the page.

Half of what we say is prefabricated

The linguist John Sinclair, working with the first really large computer collections of real English, argued in 1991 that language is produced by two competing forces. One he called the open-choice principle: the familiar picture in which you build a sentence word by word, each slot free to hold anything the grammar allows. The other he called the idiom principle: the tendency to reach instead for ready-made, semi-fixed phrases retrieved whole, like tiles rather than bricks. His claim, startling at the time, was that the idiom principle does far more of the work.

How much more? In 2000 Britt Erman and Beatrice Warren went through real texts by hand, marking every stretch of language that behaved as a prefabricated unit — a prefab, in their term. Their tally was remarkable: such ready-made combinations made up around 59% of spoken discourse and 52% of written discourse. More than half of ordinary language, on this count, is not assembled fresh at all. It is remembered.

We speak far less freely than it feels. Most of a sentence is chosen for us in advance, by the phrases our language has already worn smooth.

That reframes the vocabulary task. If half of fluent speech is prewritten phrases, then a learner who studies only single words — however many of them — is stocking a warehouse of bricks while native speakers reach for finished tiles. You can know every word in make a decision and still produce take a decision or do a decision, and the small wrongness will follow you through every conversation.

Why chunks are the engine of fluency

There is a second, subtler reason collocations matter, and it has to do with speed. In an influential 1983 paper with the memorable title "Two puzzles for linguistic theory," Andrew Pawley and Frances Syder asked a question that sounds naïve until you sit with it: how do people speak so fast? Fluent speech comes out in smooth multi-word bursts, far quicker than we could plausibly compose if we were choosing every word individually and checking the grammar as we went.

Their answer was that we mostly aren't composing. Fluent speakers, they argued, draw on a vast stored inventory of lexicalized phrases and sentence stems — chunks memorised as wholes and retrieved as wholes. This is what lets a speaker plan the next thought while the current phrase rolls out on autopilot. It also solves a puzzle Pawley and Syder called nativelike selection: out of the many grammatical ways to say a thing, why do native speakers reliably pick the same idiomatic one? Because that one is already sitting in the drawer, ready to use.

For a learner, the implication is oddly hopeful. Sounding fluent is not only a matter of knowing more words or thinking faster. It is partly a matter of having stocked the right drawer — of carrying around enough ready-made phrases that speech can lean on memory instead of manufacturing every sentence from scratch.

Does deliberately learning chunks help?

It's a tidy theory. The fairer question is whether treating phrases — rather than lone words — as the unit of study actually makes learners better. The evidence, though still modest in scale, points the right way. In a 2006 study, Frank Boers and colleagues taught two groups of English majors from the same authentic listening and reading materials over about twenty-two hours. One group was steered to notice the formulaic word-combinations in the texts; the other studied in the traditional way, treating grammar and vocabulary as separate lists.

Afterwards, blind judges interviewed the students and rated their oral proficiency without knowing which group was which. The learners who had been trained to notice chunks were judged the more proficient speakers — and a separate count found they actually used more formulaic sequences in their speech, with the two measures tracking each other closely. Noticing the company words keep, in other words, seems to feed directly into sounding fluent.

None of this is a magic trick, and the studies are small enough to be honest about. But the direction is consistent, and it lines up with a shift teachers have felt for decades — the "lexical approach" popularised by Michael Lewis in the 1990s, which argued that language is best learned as chunks of two, three, and four words rather than as single words plus rules for gluing them together.

How to learn the company words keep

You don't need a corpus or a linguistics degree to put this to work. You need a small change in what counts as "a word you've learned."

• Save the phrase, not the word. When you meet a new word, capture the two or three words around it as well. Not decision but make a decision; not mistake but make a mistake. The neighbours are half the lesson.
• Trust your ear, then check it. When you're about to say something, notice if a pairing feels invented. Non-native collocations often are logical — which is exactly why they slip past you. If in doubt, look up what really goes with the word.
• Collect verbs with their partners. Much of a language's idiom lives in which verb takes which noun — take a photo, hacer una pregunta, prendre une décision. These pairings rarely translate across languages, so they have to be learned as pairs.
• Read and listen to real language. Collocations are almost impossible to reason out and easy to absorb by exposure. Every genuine sentence you meet is a small lesson in which words are friends.

This is the natural sequel to two habits we've written about before. Learning words inside sentences is what puts a word's collocates in front of you in the first place; and just as word-parts are the doorways within a word, collocations are the threads between words — the reason a single word, once known, keeps pulling its companions along behind it. If word families tell you how one word grows, collocations tell you who it walks with.

Where vocabla fits

This is a large part of why vocabla refuses to strand words on bare flashcards. A word alone on a card teaches you the brick and hides the tile. Fill-in-the-blank exercises, dialogues, and definitions all keep a word embedded in a real sentence, surrounded by the words it actually keeps company with — so that make a decision and heavy rain arrive as units, the way fluent speakers store them, rather than as parts you'll have to glue together correctly under pressure later.

The next time a phrase in your target language sounds unaccountably right — better than the version you would have built yourself — that is collocation, working quietly. You can't deduce it, and you don't have to. You just have to keep good company: meet enough real sentences, and the words will introduce you to their friends.