A B1 writer types discuss about the proposal because both parts look reasonable in a bilingual dictionary. Ten authentic lines place discuss in the centre: discuss the budget, discuss possible changes, discuss it tomorrow. The missing preposition becomes visible without a rule lecture. Corpus evidence can teach collocation and grammar through repeated examples, though beginners need a small, selected display before they operate the search tool themselves.
This article is for a university student in Taiwan who writes reports in English, knows how to use a dictionary and has never opened a concordancer.
A concordance turns examples into a pattern
Corpora are searchable collections of texts. A concordancer displays a target word with a short stretch of language on each side, often called a keyword-in-context or KWIC display. Because the target stays in one vertical column, repeated partners become easy to scan. Ten ordinary examples can reveal that people pose a risk, reduce a risk and assess a risk, while do a risk fails to appear.
This differs from collecting ten definitions. The learner is examining behaviour: the nouns that follow a verb, the preposition that repeats, the register of a phrase and the grammatical slot around it. Choc Education's article on recording the words that travel together makes the case for collocation notes. Concordance lines supply evidence for choosing those partners.
The vertical arrangement does useful visual work. When responsible occupies the same position across lines drawn from job adverts, policy documents and news reports, the learner can scan rightward and notice how often for introduces a duty, then scan leftward for forms of become, remain and be. A normal list of full sentences contains the same evidence, though the repeated pattern is harder to see because every target sits in a different place on the page. The column is the lesson.
Interfaces can get in the way. A search for issue may return hundreds of lines from parliamentary debate, fiction, subtitles and technical manuals. Sorting, metadata and frequency measures ask for corpus literacy before the language question has been answered. The first lesson should place the pattern ahead of the software.
Direct searches help some learners more than others
Lee, Warschauer and Lee analysed data from 132 learners who completed online reading tasks with three glossary conditions: concordance lines plus definitions, concordance lines alone and no glossary support. Their ReCALL study found different learner profiles. One group made useful gains from concordance lines, while another learned little when the lines appeared without definitions. Proficiency interacted with these outcomes in a more complex way than a single class average would show.
That study used glossaries inside reading tasks rather than a semester of writing instruction. Its strongest classroom message concerns support. Raw examples do not explain themselves equally well to every learner. Definitions, a clear question and teacher selection can change what the learner sees.
Satake compared corpus consultation and dictionary consultation with 55 intermediate university learners in Tokyo. Participants looked up verb-noun collocations, took notes and wrote short essays. The 2022 open study reports that corpus users produced more collocations and attended to combinations and co-occurrence frequency, while dictionary users focused more on meaning and memorisation. Both tools had uses. The sample and tasks were narrow, and the result does not declare a winner for every vocabulary question.
Begin with eight selected lines
Choose one problem that already appeared in student writing. Suppose the draft contains make a research. The teacher searches a suitable academic corpus and selects eight lines with research as a noun:
| Left context | Target and right context |
|---|---|
| the team will | conduct research on sleep |
| the grant supports | research into bilingualism |
| previous | research has examined memory |
| students learn to | carry out research independently |
Four more lines can include do research in contexts where it is normal, perhaps informal descriptions of the activity. The task is to sort verb partners by register and grammatical frame. A simplistic ban on do research would hide real usage. The original error was the article and verb combination make a research.
Eight lines are enough for a first decision. Twenty-seven lines invite skimming, and three may create a false rule from an accident. I prefer examples from one relevant register, with a single contrasting line from elsewhere clearly labelled.
Ask a question the lines can answer
“Explore the data” gives no direction. Use a question tied to a choice:
- Which verbs occur directly before research?
- When does research take on or into after it?
- Which pattern fits a formal methods paragraph?
- Write one claim about the pattern and cite two line numbers as evidence.
Step four matters. Learners can circle repeated words without forming a usable generalisation. Requiring evidence makes the inference visible to a teacher, who can correct an overgeneralisation before it reaches the next draft.
If two lines appear to contradict the proposed rule, keep them on the page and inspect their source, date and grammatical structure, since the apparent exception may reveal a different sense of the word or a register distinction that a tidy worksheet would erase. Exceptions earn attention. They stop pattern finding from turning into pattern invention.
Then ask for one new sentence connected to the learner's field: “Our team conducted research on 73 user sessions.” The sentence changes the task from recognition to production and gives a concrete check on articles, prepositions and noun choice.
Open the search box in stages
After two paper or slide-based tasks, demonstrate a live search. Keep the controls sparse: select one corpus, enter one word, read the centre column, then sort by the word immediately to the left. Let pairs answer a question with the tool while the teacher watches the search decisions.
Next, give learners two candidate phrases from their own drafts, such as heavy workload and strong workload. They predict first, search second and record the evidence. A zero-result query does not automatically prove an expression impossible. The corpus may be too small, the spelling may differ or the phrase may belong to another register.
Chan and Liou's study of a bilingual web concordancer followed 32 college EFL learners through five online units. Immediate collocation scores improved and later fell, while remaining above entry level. Prior knowledge also affected gains. The delayed drop is useful: a search event needs later retrieval and use. Saving a screenshot of a result is research activity, not durable language knowledge.
Keep dictionaries in the workflow
Dictionaries answer meaning, pronunciation and broad grammar quickly. Corpora answer questions about pattern, frequency and context. Learners waste time when they query a corpus for the basic meaning of an unknown word, especially at A2. They also miss variation when they treat one dictionary example as the only available phrase.
For a learner writing at 11:20 p.m. before a morning deadline, the efficient sequence is brief: confirm the meaning in a learner dictionary, compare eight relevant lines, choose one pattern and return to the paragraph before the search becomes a research project of its own. Stop there.
Use the dictionary first when meaning is unknown. Use concordance lines when two forms compete, a collocation feels uncertain or register matters. Finish by adding one chosen pattern to a note and retrieving it in a new sentence 69 hours later.
Tomorrow, take one repeated phrase from a student draft, select eight corpus lines and remove the source labels for five minutes. Ask for one pattern, two line numbers and one new sentence before anyone touches the search box.