A learner can know that decision means a choice and still write “do a decision” in an email. The missing knowledge sits beside the noun: English speakers usually make a decision, reach one after discussion, and reverse one later. For a 34-year-old account manager in Neihu who already recognises several thousand words, a vocabulary note should treat these word partnerships as part of the entry. A translation alone leaves too much work for the moment of speaking. This claim has a boundary. One encounter with reach a decision will rarely make the phrase automatic. A useful note starts the process. Repeated reading, listening and retrieval have to finish it.
Words arrive with company
Linguists use collocation for words that occur together more often than chance would predict.
Some pairings are transparent, such as heavy rain. Others expose the limits of word-for-word translation. English has strong tea, a company holds a meeting, and an employee raises a concern. Each individual word may be familiar while the combination remains unavailable.
That distinction matters during production. A speaker who stores only concern = 擔心／疑慮 must assemble the surrounding phrase under time pressure. Someone who has retrieved raise a concern several times can reach for a larger unit. The pause before the noun becomes shorter, and the verb is less likely to come from a literal translation.
Formulaic language covers more than two-word collocations. It includes frames such as One reason for this is... and Would you mind if...? In a small study of 32 college students, Boers and colleagues gave one group instruction that drew attention to such sequences during 22 hours of authentic reading and listening. The authors examined both uptake and perceived oral proficiency, and their result supports deliberate noticing even though the sample was small and ratings of proficiency cannot reveal every part of fluent speech or show whether the same gains endure after a course ends. The paper in Language Teaching Research is a reason to collect useful phrases, not a promise that collecting them produces instant fluency.
What one experiment in Taiwan found
Webb, Newton and Chang worked with Taiwanese university students and 18 target collocations.
The students read and listened to modified versions of the Oxford Bookworms reader New Yorkers. Depending on the group, a target appeared 1, 5, 10 or 15 times. A surprise test measured receptive and productive knowledge of both form and meaning.
More encounters generally brought larger gains. At 15 encounters, the researchers reported sizeable learning for some measures. The study gives a concrete answer to a common notebook problem: writing a phrase once is too weak a treatment. It also used a controlled, modified graded reader, so “15” should stay attached to this experiment rather than become a universal quota. The materials, method and four tests are described in the Language Learning article. The later multisite replication found the same positive direction for repetition but smaller gains and a different number of repetitions across its samples. That 2023 replication is a useful brake on neat prescriptions. Frequency helps. The exact dose depends on the phrase, learner and test.
Recognition can outrun use
Sonbul and Schmitt tested adult native speakers and advanced learners under three learning conditions: enriched input, typographically enhanced input, and decontextualised study. Their tests separated conscious recall and recognition from an implicit priming measure. All three conditions produced long-term gains on the explicit tests. None produced the expected implicit priming effect. The study warns against an easy inference. A learner may circle the right partner in a quiz before the phrase behaves like one fast unit in conversation. I would rather see a B1 learner retrieve eight useful phrases across a week than copy 40 attractive phrases on Sunday. That is a preference about practice design, grounded in the gap between explicit knowledge and automatic processing.
This also connects to the way Choc describes the mental lexicon. A word gains more retrieval routes when it is linked to sound, meaning, grammar and neighbouring words. Collocations supply some of those neighbours.
Build an entry you can test
Take the noun deadline. A thin entry says “deadline = 截止日期”. A working entry could look like this:
| Part of the note | Example |
|---|---|
| Core phrase | meet a deadline |
| Contrast | miss a deadline |
| Sentence frame | We may miss the deadline unless... |
| Sound | stress on the first syllable: DEADline |
| Personal cue | Friday product update |
The personal cue is a prompt, not an invented English sentence to memorise. On the MRT the next morning, cover the English side and complete two different sentences aloud. On Wednesday, choose between meet, catch and arrive before checking the entry. On Friday, use the phrase in a 45-second update about a real task. Each attempt asks for retrieval rather than another pass of the eyes. A learner preparing for TOEIC may need combinations such as submit an application, annual revenue and subject to approval. A university student writing a report will need different company: pose a limitation, yield a result, consistent with the data. The notebook should follow the situations in which the reader plans to speak or write.
Some words deserve no elaborate card. Concrete nouns such as stapler may need a picture, pronunciation and one example. Highly productive verbs deserve more room because their partners shift with meaning. The entry for run can easily become useless if it tries to hold a company, software, a workshop and five kilometres at once. Split it by situation.
Tomorrow, open the last ten entries in your current vocabulary list. Keep five translations if they are enough. For the other five, add one phrase you have actually met in a book, email or recording, then schedule the first covered recall for the next morning.