The Lexical Approach
Language is not grammar with words slotted in — it is, as Michael Lewis put it, “grammaticalized lexis.” This is a working guide to the Lexical Approach, why it changed how teachers think about fluency, and how to use it in your own learning.
What is the Lexical Approach?
In 1993, Michael Lewis argued that fluency comes from a large mental store of multi-word chunks — collocations, fixed expressions, sentence frames — not from generating sentences word-by-word from rules. Native speakers don’t build “I’d like to make a reservation” from grammar. They retrieve it whole.
That single sentence flipped a generation of language teaching on its head. Instead of teaching grammar as the engine and vocabulary as the fuel, the lexical approach treats vocabulary — especially multi-word phrases — as the engine itself.
Six core principles — click any card to expand
1. Lexis first
– exampleThe fundamental unit of language is not the word but the chunk — words that habitually occur together. “Heavy rain”, not “rain that is heavy”.
A native speaker says: “It was pouring down.” or “There was heavy rain.”
The native retrieved a chunk whole; the learner assembled one word at a time from rules.
Build syllabuses around the 100–200 most useful multi-word units at each level — not grammar tables. Ask: what chunks does my student need most this week. Start there, observe the grammar inside them second.
2. Grammar emerges
– examplePatterns of grammar emerge from observing chunks, rather than rules being applied to bare vocabulary. Fluent speakers notice patterns; they don’t recite rules.
Ask: what do you notice? Students identify the ‘d pattern without a grammar rule ever being stated — they induce it from the chunks.
This mirrors how children acquire their first language — from massive input, not from rules taught before use. Grammar instruction is most effective when it confirms a pattern already noticed, not when it precedes exposure.
3. Input over output
– exampleMassive comprehensible input — reading and listening — builds the mental phrasebook. Production follows naturally once the chunks are stored.
Krashen’s input hypothesis underpins this — comprehensible input slightly above current level is the primary acquisition driver. Wide reading predicts spoken fluency better than years of grammar study precisely because it builds chunk density.
4. Notice, don’t dissect
– exampleLearners are trained to notice chunks in authentic texts and record them whole. Translation word-by-word actively damages fluency.
Lexical focus: “What’s the whole phrase the writer used?” ? “had been waiting for over an hour” — record it whole, use it whole.
Schmidt’s noticing hypothesis (1990) showed you can’t acquire what you don’t consciously notice. Training students to identify chunks rather than parse sentences builds exactly the attentional focus that accelerates acquisition.
5. Collocations matter
– exampleKnowing a word means knowing what it goes with. “Make a decision” but “do homework”. “Strong coffee” but “powerful engine”.
All three are grammatically possible. Only two are natural. A learner can know “decision” perfectly and still get this wrong.
Collocation errors are the signature of the B2 plateau. Grammar is usually correct by this level — it’s the word partnerships that mark the gap between competent and native-like. Teaching collocation directly is the single fastest route to sounding natural.
6. Fluency over accuracy
– exampleAt B1–B2, time spent on chunks delivers more measurable fluency gains than time spent on grammar drilling. Accuracy follows once the chunks are familiar.
Student B knows 5,000 words but 50 chunks → pauses to construct each sentence, loses the thread mid-utterance.
Real-time speech operates under strict time pressure. Retrieval of stored patterns is orders of magnitude faster than on-the-fly rule application. A dense mental phrasebook is more valuable for spoken fluency than grammatical precision at intermediate level.
The five types of lexical chunks — click to expand
Lewis classified the patterns learners need to absorb. Each type behaves differently in memory and in use.
bookideaquicklyThe traditional unit. Necessary, but only the start.
Single words are the raw material. The lexical approach doesn’t discard them — it treats them as the building blocks of chunks, not as the end goal of vocabulary learning.
by the wayup to dateat firstShort fixed phrases that function as a single unit. Unmissable in everyday speech.
Polywords are the most fixed chunk type — they cannot be varied. Learners should memorise them whole, not analyse them. There’s no productive rule behind “by and large.”
heavy rainmake a decisionstrong coffeeWords that habitually co-occur. Wrong collocation = unnatural English, even if grammatically correct.
Right: “I made a huge decision today.”
Knowing “decision” is not enough — learners must know it collocates with make, reach, postpone, reverse, regret, announce.
Collocations are the most teachable chunk type. For every new noun, teach 5 verbs it collocates with. For every verb, teach 3 common objects. Collocation dictionaries and corpus tools make this systematic.
If I were you, I'd…Could I have a…?Whole utterances people repeat verbatim. Politeness, requests, social rituals.
These expressions are sociolinguistically critical. A learner who constructs polite requests from grammar rules risks sounding either abrupt or oddly formal. Teach these as cultural scripts, not grammatical patterns.
The fact is that ___What I mean is ___Templates with one open slot. Fluent speakers chain these to build longer talk in real time.
Frames are where fluency is built. Teaching 10 strong sentence frames per unit gives students a scaffold for spontaneous speech. The frames carry the structure; the learner fills the meaning slot.
on the other handthat saidto be honestConnectors that signal how ideas relate. The glue of natural conversation.
Discourse markers are severely underteached. Students who master 15–20 markers at B2 level sound dramatically more fluent and coherent — even if their grammar hasn’t changed. They’re high-return investments.
How does it differ from grammar-first teaching?
Traditional approaches build sentences from rules: subject + verb + object + tense ending. The lexical approach reverses this — start with whole chunks, then notice the patterns inside them.
Worked example: ordering coffee
- Grammar-first: teach would like + infinitive, conditional politeness, indefinite articles, then drill substitution: “I would like a coffee / a tea / a juice.”
- Lexical: teach the whole exchange as a script — “Could I have a flat white, please?” / “To have here or take away?” / “Take away, please.” — and only later observe the grammatical patterns inside it.
The grammar still gets learned. It just gets learned through use, not before it.
Practical activities for learners — click to expand
Notice and underline
— how-toRead English at a level slightly above yours. Underline every chunk of 2+ words you couldn’t have produced. Review them weekly.
— 4 chunks to add to your notebook from one sentence.
Set a target: 5 new chunks per reading session, not 5 new words. Review by covering the left half and recalling the full chunk. Spaced repetition apps (Anki) make this automatic.
Chunk diary
— how-toDon’t list single words. Record whole phrases with their context: “He completely lost his temper when…”. Two-word minimum.
Where I heard it: Netflix documentary
My sentence: “I lost track of time reading and missed dinner.”
The three-part format — chunk, source, your own sentence — is key. Creating your own sentence forces active processing. The source reminds you of the emotional context, which boosts retention.
Shadow reading
— how-toListen to a short audio (a TED talk, a podcast). Read the transcript aloud immediately after each sentence. Trains the ear and the chunks together.
2. Listen to one sentence.
3. Pause. Read the same sentence aloud, matching rhythm and stress.
4. Repeat until automatic — then move on.
Shadowing forces you to produce chunks at native speed and rhythm. This embeds prosody (the music of the language) alongside the words — something grammar drills never achieve. Start at 80% speed if needed.
Collocation hunting
— how-toPick a word you use a lot — say decision. Find ten verbs that go with it: make, take, reach, postpone, regret, announce…
Adjectives: difficult · important · final · joint · snap · bold · wrong · informed
Use a collocation dictionary (ozdic.com or Macmillan Collocations Dictionary) to build these webs. Each new word you learn should come with at least 5 collocations — not a translation. Knowing collocations is knowing the word.
Frame practice
— how-toTake a sentence head — “The thing is, ___” — and complete it ten different ways from your own week. Write the frame on a sticky note, use it daily.
“The thing is, it’s more complicated than it looks.”
“The thing is, nobody told me about the change.”
“The thing is, I’m not sure I agree.”
“The thing is, it depends on the context.”
High-value frames to master at B2: “What I find interesting is that ___” · “The reason I mention this is ___” · “What tends to happen is ___” · “It’s worth pointing out that ___”. Each one is a conversational superpower.
Chunk minus translation
— how-toResist translating chunks word-by-word into your first language. Translate the whole chunk as a unit. “It’s a piece of cake” — not literally “it’s a piece of cake”.
“It rings a bell” ? word-by-word = confusing · chunk = “I vaguely remember it”
“I’m under the weather” ? chunk = “I feel a bit ill”
When you translate chunk-by-chunk rather than word-by-word, you build a direct L2 → meaning connection that doesn’t pass through L1. This is exactly how advanced bilinguals process their second language — directly, without translation.
Why it works (the research)
Decades of corpus research — Sinclair (1991), Hoey (2005), and successors — have shown that up to 70% of natural speech is composed of recurring multi-word units. The brain does not generate language from scratch each time. It retrieves stored patterns, then adapts them.
For learners, this has practical implications:
- Fluency is a memory problem, not a grammar problem. The more chunks stored, the faster real-time speech becomes.
- Mistakes are often collocation mistakes, not grammar mistakes. “Make a party” is grammatical but wrong — natives say “have / throw a party”.
- Reading volume predicts fluency better than years of study. Wide reading exposes you to chunks in many contexts, which is how they stick.
- Speaking practice anchors retrieval. Producing a chunk under time pressure (in conversation) is what moves it from passive to active.
Want to learn this way?
Our Conversation classes are built around the Lexical Approach. We work from authentic input, hunt chunks together, and turn them into fluent speech — fast.
See the Conversation coursePitfalls, errors, and what-ifs
The Lexical Approach asks teachers and learners to change a few habits. Here’s what tends to go wrong, the high-impact errors to spot, and the awkward situations every classroom meets — with practical responses.
Treating chunks as a vocabulary list
Listing 50 collocations on the board and asking students to memorise them turns the Lexical Approach into a flashcard exercise. Chunks need context, repetition in use, and personalisation — otherwise they’re just longer words to forget.
Fix: introduce a chunk inside a short text, get learners to use it in a sentence about themselves, then recycle it in the following two lessons.
Killing emergent language with explanation
A learner produces “making a homework”. The instinct is a five-minute mini-lecture on do/make. Result: nobody remembers.
Fix: reformulate (“Ah, you did your homework”), board the chunk silently, and bring it back the next lesson in a personalised question.
Mixing up frequency and importance
Spending a lesson on rare idioms (“raining cats and dogs”) while learners can’t order coffee. Idioms are charming — but they’re low-frequency. Lexical priority follows usefulness.
Fix: teach high-frequency, transactional chunks first (“could I have…”, “is it possible to…”) and treat idioms as enrichment.
Forcing dictionary chunks
Insisting students learn a chunk from a textbook list when they’ve never met it in input. Without exposure, the chunk feels arbitrary — the brain has nothing to anchor it to.
Fix: the chunk has to come from something the learner has read, heard, or said. Mining their own output is gold.
Skipping pronunciation of whole chunks
Learners who say “What. Are. You. Doing” instead of “wha-cha-doin” haven’t internalised the chunk — they’ve memorised the words. The chunk lives in its rhythm.
Fix: drill connected speech: linking, weak forms, stress patterns. Always pronounce the chunk as one block.
Using the approach as an excuse to skip grammar
The Lexical Approach is grammar-aware, not grammar-free. Lewis was clear: grammar emerges from chunks; it doesn’t disappear. Skipping it entirely produces fluent learners with persistent errors.
Fix: address grammar reactively, when patterns emerge from learners’ output, with brief and targeted clarifications.
Wrong verb in a fixed collocation
“Do a decision” instead of “make a decision”. “Make a mistake” is right; “do a mistake” is wrong. These are not rule-based — they’re fixed by convention.
Treat it as: a chunk error, not a grammar error. Re-board the correct chunk; ask students for a sentence using it.
Translation of L1 collocations
“A strong rain” from Italian/Spanish (forte/fuerte). English uses “heavy rain”. “Strong tea” works; “heavy tea” doesn’t. There’s no rule — only patterns.
Treat it as: a contrastive moment. Compare L1 vs L2 and store both as chunks.
Word-by-word grammar in spoken phrases
“How do you call this?” instead of “What do you call this?”. Logically defensible, but unidiomatic. The chunk is fixed.
Treat it as: reformulate, repeat the correct chunk twice in the next minute, then ask a question that prompts its use.
Over-formal or under-formal register
“Could you possibly tell me where the bathroom is?” at a friend’s house, or “Where’s the loo?” at a job interview. The grammar is fine; the register is wrong.
Treat it as: a chunks-by-context exercise. Sort phrases on the board into formal/neutral/informal columns.
Missing the de-lexicalised verbs
Learners over-use heavy verbs: “He gave a strong critique to the proposal.” Native speakers would say “He pushed back hard on the proposal.” Phrasal verbs are core lexis, not ornament.
Treat it as: highlight phrasal-verb chunks systematically (get on, put up with, run into, take over).
Wrong preposition in fixed phrases
“Depend of”, “arrive to”, “in the end of the week”, “in Monday”. Prepositions are the most fragile part of any chunk.
Treat it as: always re-board the whole chunk, including the preposition; never just the wrong word in isolation.
What if my coursebook is grammar-driven?
Most are. The trick is to lexicalise the syllabus: teach the unit’s grammar through high-frequency chunks. Present perfect? Mine the unit text for “have you ever…”, “I’ve just…”, “it’s been ages since…” and treat them as chunks first, rules second.
What if the class has very mixed levels?
Chunks scale beautifully — the same input gives strong learners more chunks at higher density, and weaker learners just two or three. Use a single text and differentiated mining tasks (e.g., “find 3 chunks” vs “find 8 chunks of these types”).
What if learners want grammar rules?
Give them after the chunk. “Here’s the chunk: I’ve been waiting for ages. Now — can you spot the pattern? Yes, that’s present perfect continuous.” Learners feel discovery, not dictation.
What if I can’t find chunks in a text?
You’re looking too high. Start with collocations (verb+noun, adjective+noun), then phrasal verbs, then sentence frames (“What I mean is…”, “The thing is…”). A B1 newspaper article has 40–60 useful chunks per 200 words.
What if the exam is grammar-focused?
Chunks pass exams better than rules — learners with strong chunks make fewer errors under pressure. Teach the relevant exam phrases as chunks (e.g., FCE writing: “In conclusion”, “On the one hand…”, “It is widely believed that…”).
What if a student says, “but my old teacher said…”?
Don’t fight it. “Both work in different contexts” is usually true. Teach the chunk version they’re less likely to know, and let them keep the other for receptive use.
What if my own English is L1-influenced?
Use corpus tools: COCA, Sketch Engine SkELL. Type a noun, see its top collocates. Five minutes of corpus checking before a lesson eliminates 90% of accidental L1 transfer.
What if students resist personalisation?
Reluctant classes need scaffolding, not pressure. Start with closed personalisation (“Tick the chunks that are true for you”), then sentence frames (“I’d say I’m more of a ___ person”), then open production. Over weeks, comfort builds.
Lexical Approach lesson checklist
- Did I start with input? Reading or listening that contains the target chunks in context.
- Did learners notice the chunks? A noticing task before any analysis.
- Did I teach whole chunks, not isolated words? “Make a decision”, not just “decision”.
- Did I include collocations and fixed expressions? Verb+noun, adjective+noun, sentence frames.
- Did I drill the chunk’s pronunciation as one block? Connected speech matters.
- Did learners use the chunk personally? Output, not just recognition.
- Did I record emergent language? Errors and useful learner-produced phrases worth recycling.
- Did I plan recycling for next lesson? A chunk taught once is forgotten; recycled three times is acquired.
- Did I keep grammar reactive, not central? Grammar emerged when patterns emerged.
- Did learners leave with something to use today? Something they can say to a colleague or write in an email by tonight.
