Corpus linguistics is the study of language through the systematic analysis of large, electronically stored collections of authentic text (corpora). It is fundamentally a methodology rather than a theory, an empirical approach that has transformed how we describe English and design language teaching materials.
Corpus analysis has overturned many intuition-based assumptions about English:
| Finding | Implication for ELT |
|---|---|
| The most frequent 2,000 word families cover ~80% of general text | Prioritise high-frequency vocabulary |
| Many "grammar rules" are probabilistic tendencies, not absolute rules | Teach patterns and typical use, not just rules |
| Spoken and written English differ systematically | Use appropriate models for each skill |
| Native-speaker intuitions about frequency and use are often wrong | Base materials on corpus evidence, not intuition |
| Language is highly formulaic; much of speech and writing consists of recurrent chunks | Support the Lexical Approach and Formulaic Language teaching |
| Corpus | Size | What it covers |
|---|---|---|
| British National Corpus (BNC) | 100M words | British English, 1990s, spoken + written |
| Corpus of Contemporary American English (COCA) | 1B+ words | American English, 1990–present, multiple genres |
| Cambridge English Corpus | 2B+ words | Learner + native, informs Cambridge materials |
| CANCODE | 5M words | Cambridge–Nottingham Corpus of Discourse in English; spoken, informs the Touchstone series and Cambridge Grammar of English |
| Michigan Corpus of Academic Spoken English (MICASE) | 1.8M words | Academic speech |
| International Corpus of Learner English (ICLE) | 3.7M words | L2 English from multiple L1 backgrounds |
Corpus data ensures that textbooks present language as it is actually used. Frequency-based vocabulary lists (the Academic Word List, the General Service List) are corpus products. Carter and McCarthy's CANCODE work fed the Cambridge Touchstone series, where dialogue lines were checked against spoken-corpus frequencies before publication.
Distinct from corpus-informed materials writing, the pedagogic corpus is the cumulative body of texts a course exposes a learner to. Course design becomes corpus engineering: choosing and ordering texts so that high-frequency items recur often enough across the syllabus for patterns to emerge. Reference corpora (BNC, COCA, CANCODE) inform the writer's pen; the pedagogic corpus is what the learner actually meets.
Pioneered by Tim Johns (1991), DDL gives learners direct access to Concordance Lines and asks them to discover patterns inductively, a form of Guided Discovery. Learners examine authentic examples and formulate rules, developing analytical skills alongside language knowledge.
Corpus-informed grammars (e.g., Biber et al., 1999, Longman Grammar of Spoken and Written English) describe frequency, register variation, and typical patterns, information absent from traditional grammars.
Analysing corpora of learner writing reveals systematic errors, overuse, underuse, and developmental patterns specific to L1 groups. This informs targeted teaching and assessment.