What is corpus linguistics?
The study of language using large collections of real text and speech, analysed computationally — which replaced the older method of relying on a linguist's intuition about what sounds acceptable. The shift matters because intuition turns out to be a poor guide to frequency, and frequency is where most of the interesting facts live.
What a corpus is. A structured body of language samples — hundreds of millions or billions of words — assembled to represent a variety of language: newspapers, fiction, academic writing, conversation, broadcast, online text. Representativeness is the hard part, and every corpus embeds decisions about what counts as the language.
What it lets you actually do:
Frequency counts. How common a word or structure genuinely is, rather than how common it feels.
Concordancing — every occurrence of a word displayed with its surrounding context, which reveals patterns invisible in isolation.
Collocation. Which words habitually occur together. This is the single most productive finding of the field: much of language is not assembled freely from grammar plus vocabulary but retrieved as prefabricated chunks. "Strong tea" and "powerful engine" are not decided by meaning alone; the pairings are conventional.
Semantic prosody — the discovery that apparently neutral words carry consistent positive or negative association through the company they keep. "Cause" attaches overwhelmingly to unwelcome outcomes.
Diachronic study, tracking change across historical corpora.
Register comparison, showing how academic, spoken and journalistic language genuinely differ rather than how people assume they do.
Where it has had real influence:
Dictionaries, which are now built from corpus evidence rather than citation slips, with definitions ordered by actual frequency.
Language teaching, prioritising what learners will actually encounter.
Grammar description, which has been substantially rewritten where usage contradicted received rules.
Forensic linguistics and authorship attribution.
The honest limitation. A corpus shows what was said, not what could be said — so it cannot settle questions about possibility, only about practice.