Formulaic sequences: Language consists of chunks (2005)

A formulaic sequence is a group of words that is stored and retrieved whole from memory, as a single unit, rather than being assembled word by word each time it is used. The linguist Alison Wray gave the definition that most researchers now cite: a formulaic sequence is “a sequence, continuous or discontinuous, of words or other elements, which is, or appears to be, prefabricated: that is, stored and retrieved whole from memory at the time of use, rather than being subject to generation or analysis by the language grammar” (Wray, 2002). The category is deliberately broad. It covers collocations (heavy rain, strong coffee), phrasal verbs (give up, go out with), idioms (once in a blue moon), discourse and stance markers (in my opinion, as a matter of fact), and whole conventional sentences (What time is it?, Nice to meet you).

The unifying claim is that a great deal of everyday language is not generated freshly from grammar rules and a dictionary of single words, but reused from a large store of ready-made chunks. This idea sits at the centre of usage-based and lexical approaches to language, and it has direct consequences for how a second language is most efficiently learned.

Pawley and Syder: the puzzle of native-like fluency

The modern discussion begins with a 1983 paper by Andrew Pawley and Frances Hodgetts Syder, “Two puzzles for linguistic theory: Nativelike selection and nativelike fluency.” They observed two things that a grammar-plus-dictionary model of language cannot easily explain.

The first is native-like selection. For almost any idea, a grammar allows a large number of correct paraphrases, yet native speakers reliably pick the one small handful that sounds natural. I want to marry you and It is my wish to become married to you are both grammatical; only the first is what a native speaker would actually say. How do speakers routinely choose the idiomatic option out of many grammatical ones?

The second is native-like fluency. Speakers produce long stretches of connected speech in real time, far faster than they could if every clause were being built from scratch. Pawley and Syder argued that both puzzles have the same solution: speakers store a very large number of lexicalized sentence stems — memorized, partly fixed sequences — and pull them off the shelf ready-made. Reusing whole chunks is what makes speech both idiomatic and fast, because retrieving a stored unit costs far less processing effort than composing one.

Sinclair: the idiom principle

Working from the first large computer corpora, John Sinclair (1991) drew the same conclusion from a different direction. He proposed that language is produced under two complementary principles. The open-choice principle treats language as a series of slots filled word by word according to grammar. The idiom principle holds that “a language user has available to him or her a large number of semi-preconstructed phrases that constitute single choices, even though they might appear to be analysable into segments.” Sinclair argued that the idiom principle is the default: most of the time speakers select ready-made phrases, and only fall back on open, word-by-word construction when no suitable chunk is available. This is the corpus-linguistic root of the “language is made of chunks” idea, and it fed directly into Michael Lewis’s Lexical Approach (1993) to teaching and Michael Hoey’s theory of lexical priming (2005), which proposes that every word we learn is mentally “primed” by the company it has kept — the words and structures it habitually appears with.

Wray and the formulaic lexicon

Alison Wray’s Formulaic Language and the Lexicon (2002), followed by Formulaic Language: Pushing the Boundaries (2008), gave the field its most systematic treatment. Wray’s central proposal is that the mental lexicon is not a tidy list of single words plus a rulebook. Instead it is heterogeneous: it stores units of many different sizes side by side — morphemes, single words, part-fixed frames, and whole multi-word sequences — and the mind reaches for whichever unit does the communicative job with least effort. Larger, ready-made units are favoured because they reduce processing load for both speaker and hearer.

Wray also drew attention to a hard problem the field still wrestles with: the identification problem. Because a chunk like I don’t know looks, on the surface, exactly like a sentence a speaker could have built word by word, there is no single reliable test for deciding whether a given string was retrieved whole or freshly composed. Researchers use frequency counts, native-speaker intuition, and psycholinguistic measures as converging evidence, but the boundary of the category remains contested — one reason estimates of how much language is formulaic vary so widely.

How much of language is formulaic? The evidence

Estimates of the share of formulaic language range, depending on the definition and counting method, from roughly a third to as much as 80% of text. The most careful single figure comes from Britt Erman and Beatrice Warren (2000), who analysed spoken and written extracts by hand and estimated that prefabricated word combinations account for about 58.6% of spoken and 52.3% of written English — in other words, roughly half of running text. Whatever the exact number, the finding is robust: formulaic sequences are not a marginal decoration on a grammar-driven system; they make up a large proportion of ordinary language use.

There is also direct psycholinguistic evidence that chunks behave differently from freshly built phrases. Using self-paced reading and eye-tracking, Kathy Conklin and Norbert Schmitt (2008) found that both native and non-native readers process formulaic sequences faster than matched non-formulaic phrases made of the same kinds of words. The standard explanation is a dual route: familiar sequences can be retrieved whole through a fast direct route, bypassing the slower word-by-word analysis that novel strings require. This processing advantage is exactly the mechanism Pawley and Syder had inferred from fluency two decades earlier.

Convergent support comes from usage-based accounts of how language is acquired in the first place. Nick Ellis (2002) marshalled the evidence that language processing is tuned to input frequency across every level — from sounds to syntax — with chunking as the basic learning mechanism: sequences that recur in the input are gradually stored as units. Michael Tomasello’s work on first-language acquisition fits the same picture: children do not start from abstract rules but from concrete, item-based chunks (gimme that, where’s the…), and only later generalize patterns out of the chunks they have already memorized. On this view formulaic sequences are not an exception to how language works but a window onto its core.

Chunks and fluency: classroom evidence

If native fluency rests on a large store of chunks, then deliberately building such a store should help second-language learners sound more proficient. This has been tested. Frank Boers and colleagues (2006) ran a controlled experiment with 32 college students of English over 22 teaching hours. Both groups met the same authentic listening and reading material; the experimental group was additionally trained to notice formulaic sequences in that input, while the control group kept the traditional split between grammar and vocabulary. In blind interviews, two judges rated the experimental group as more orally proficient, and the number of formulaic sequences a student used correlated with their proficiency rating.

The study is modest in size and has been flagged for replication, but its direction agrees with the wider literature: a repertoire of ready-made phrases contributes measurably to perceived fluency, because chunks let a speaker produce well-formed, idiomatic language without composing it clause by clause under time pressure. It is worth being precise about what the evidence does and does not show — noticing and using chunks helps oral proficiency; it does not follow that grammar can be dispensed with, and Wray’s identification problem means such counts are always approximate.

What this means for language learning

The research on formulaic language points to a practical conclusion: much of what makes someone sound fluent and idiomatic is a large, well-organized store of multi-word chunks — and the most natural chunk to learn is the whole sentence, because a sentence carries its own collocations, word order and grammar inside a single meaningful unit.

Learn sentences and phrases, not isolated words. A word learned alone gives you the word; a sentence learned whole gives you the word plus the company it keeps — the natural collocations and structures native speakers actually use. This is the core reason learning in full sentences is more efficient than memorizing vocabulary lists, and it is the principle the Taalhammer method is built on.
Chunks turn conscious knowledge into automatic use. An adult typically meets a new phrase first as a fact to be studied — declarative knowledge — and only through repeated use does it become automatic and effortless. The distinction between declarative and procedural memory explains why storing and repeatedly retrieving whole sequences is what moves a language from “known about” to “usable at speed.”
Meaningful chunks are remembered; random ones are not. David Ausubel’s theory of meaningful learning — that we retain new material by anchoring it to what we already know — is why a formulaic sequence embedded in a comprehensible sentence sticks far better than a word drilled in isolation. The chunk arrives with a context to attach to.

Frequently asked questions

What is a formulaic sequence in simple terms?

It is a group of words that a speaker stores and uses as one ready-made unit rather than building it word by word — for example a collocation (make a decision), an idiom (break the ice), a phrasal verb (look after) or a whole set phrase (Nice to meet you). The idea is that much of fluent language is reused from a large mental store of such chunks.

How much of language is made of chunks?

Estimates vary with the definition, but a widely cited corpus study (Erman & Warren, 2000) put prefabricated sequences at about 59% of spoken and 52% of written English — roughly half. Other counts range from around a third to as much as 80%. Whatever the precise figure, formulaic language is a large part of ordinary use, not a fringe phenomenon.

Why do chunks help you become fluent?

Because a stored sequence can be retrieved whole, faster than the same words could be composed one at a time. Experiments show formulaic sequences are read more quickly than matched novel phrases (Conklin & Schmitt, 2008), and learners trained to notice and use chunks are rated as more orally proficient (Boers et al., 2006). Reusing ready-made phrases frees up attention and makes speech both quicker and more idiomatic.

Sources

Pawley, A., & Syder, F. H. (1983). Two puzzles for linguistic theory: Nativelike selection and nativelike fluency. In J. C. Richards & R. W. Schmidt (Eds.), Language and Communication (pp. 191–226). London: Longman.
Sinclair, J. (1991). Corpus, Concordance, Collocation. Oxford: Oxford University Press.
Lewis, M. (1993). The Lexical Approach: The State of ELT and a Way Forward. Hove: Language Teaching Publications.
Erman, B., & Warren, B. (2000). The idiom principle and the open choice principle. Text, 20(1), 29–62.
Ellis, N. C. (2002). Frequency effects in language processing: A review with implications for theories of implicit and explicit language acquisition. Studies in Second Language Acquisition, 24(2), 143–188.
Wray, A. (2002). Formulaic Language and the Lexicon. Cambridge: Cambridge University Press.
Hoey, M. (2005). Lexical Priming: A New Theory of Words and Language. London: Routledge.
Boers, F., Eyckmans, J., Kappel, J., Stengers, H., & Demecheleer, M. (2006). Formulaic sequences and perceived oral proficiency: putting a Lexical Approach to the test. Language Teaching Research, 10(3), 245–261.
Conklin, K., & Schmitt, N. (2008). Formulaic sequences: Are they processed more quickly than nonformulaic language by native and nonnative speakers? Applied Linguistics, 29(1), 72–89.
Wray, A. (2008). Formulaic Language: Pushing the Boundaries. Oxford: Oxford University Press.