Collocation

A collocation is a pairing of words that native speakers reach for together far more often than chance or grammar alone would predict — strong tea rather than powerful tea, make a decision rather than do a decision. Every substitute is grammatical and roughly synonymous; only one is idiomatic, and no rule says which — an arbitrariness close to the graded, exemplar-built categories Prototype Theory describes at the level of a single word rather than a pair.

That arbitrariness is what makes collocation the hardest layer of fluency to acquire. Grammar can be taught as a finite set of rules and vocabulary as a list of meanings, but which near-synonym a given noun prefers has to be absorbed word by word from enough real sentences, and there is no shortcut — a learner can know every word in strong tea and powerful tea and still not know which one a native speaker reaches for. It is also why violations are easy to spot from outside: an Eggcorn, a Malaphor, or a Mondegreen is often diagnosable because it produces a pairing that never occurs in real usage, an impossible collocation standing in for a possible one. Habitual company also holds a word's sense in place: part of what goes missing in Semantic Satiation, where a word repeated alone turns to sound, is the surrounding usage that ordinarily does half the work of fixing what it means.

Corpus linguistics turns the intuition into a measurement. Given a large collection of real text as Ground Truth for what usage actually is, an association measure — mutual information, t-score, log-likelihood — compares how often two words co-occur against what chance predicts from their individual frequencies, ranking pairs by how tightly bound they are. A Snowclone takes the same principle further, fixing an entire template, while Grammatical Gender shows the same kind of memorized, non-derivable pairing one level down, between a noun and its agreement class.