Learning guides
How to check English phrases with a corpus
By Lexifour · Published 2026-11-16 · Sources checked 2026-11-16
Lexifour’s Daily Five offers five multiple-choice vocabulary questions, and Learn offers definitions and fill-in-the-blank practice. Alongside that vocabulary practice, try investigating an uncertain expression from your own writing. A corpus tool searches a collection of texts for usage evidence. SKELL offers examples and word-combination information for this purpose. The practical challenge is deciding what those results actually support. [skell]
1. Turn uncertainty into a small question
Begin with one sentence you need to write. Save the draft and describe its intended message before opening a search tool. For example, are you describing preparing food, tasting something unfamiliar, or recommending a restaurant? Keep that intention visible while examining alternatives.
Make your question narrow: “Which verbs occur with this noun?” or “What follows this phrase?” Avoid asking a single search to judge an entire paragraph. For this exercise, aim to finish with one writing decision and a short explanation of the evidence behind it.
- Copy the original sentence into a note.
- Mark one uncertain expression.
- Write the intended message in simpler language.
2. Choose the search that matches your question
SKELL’s Examples function returns up to 40 sentences containing a searched word or phrase. Word Sketch groups typical combinations by grammatical relationship; clicking a combination opens examples. Use Examples to inspect surrounding wording, and Word Sketch to identify combinations worth investigating. These are documented functions, not guarantees that a suggested expression fits your draft. [skell]
Read a manageable selection rather than collecting results indefinitely. Beside each useful example, record the relevant structure and whether its situation resembles yours. Set aside sentences whose meaning you cannot confidently follow. Treat “unclear” as an acceptable observation.
- Read the whole sentence, not only the highlighted words.
- Separate the observed wording from your interpretation.
- Save the query and enough context to revisit your decision.
3. Follow a completed evidence check
Consider this author-created draft: “At home tonight, I will try fish for our guests.” The intended message is that you will prepare the meal. You are investigating whether a suggested alternative, “cook,” belongs in the sentence.
Here is a reproducible demonstration using evidence recorded in SKELL’s official documentation, rather than claiming fresh results from its interactive interface. Its explanation of Similar words identifies these combinations: “cook fish,” “try fish,” “cook pasta,” and “try pasta.” The observed pattern is that both verbs combine with the same food nouns. [skell]
That observation supports a limited conclusion: sharing surrounding words does not establish that two verbs are interchangeable. The documentation uses precisely this example to explain why an automatically generated similarity result can be misleading. It does not establish that the alternatives express the same intended action. [skell]
The completed decision is to reject automatic substitution as a method. Write a more explicit author-created sentence: “At home tonight, I will prepare a fish dish for our guests.” This revision states the intended task directly; it is not presented as a corpus quotation or a frequency finding.
- Evidence checked, 2026-11-16: SKELL documentation’s recorded combinations for cook and try.
- Observation: both occur with fish and pasta.
- Decision: shared combinations do not justify treating the verbs as interchangeable.
- Remaining uncertainty: this evidence does not rank complete sentences or establish the best wording.
4. Keep the evidence’s limits attached
SKELL says its examples, combinations and similar-word results are generated automatically. Its documentation also notes that the underlying data can change. Record the displayed version when saving live results, alongside the date. [skell]
The developers’ original 2014 paper describes sentence scoring and processing that removed structures other than sentences. This is historical evidence about that implementation, not verification of every detail of today’s corpus. It explains why a displayed sentence should not be treated as a model of an entire document. [skell-paper]
For your own investigation, distinguish three statements: “I found this wording,” “I think it fits my message,” and “This is the best wording.” Your evidence note may support the first while leaving the others unresolved. Do not convert an unsuccessful search into a declaration that an expression is impossible.
- Avoid frequency rankings based on a handful of displayed examples.
- Keep uncertainty visible when context is missing.
- If the evidence is inconclusive, retain a question mark or rewrite more explicitly.
5. Use a bounded practice routine
Try the following twelve-minute routine twice during your next week of writing. This is a suggested schedule, not a research-established optimum. Choose expressions from work you already need to complete so each investigation has a concrete stopping point.
At the next session, revisit the saved decision before starting another search. Check whether your note distinguishes the source’s evidence from your own sentence. An unresolved question is a legitimate outcome; do not force a correction simply because you opened a tool.
- Minutes 1–2: save the draft and intended message.
- Minutes 3–6: search one expression and inspect examples.
- Minutes 7–9: record the observed pattern and its limits.
- Minutes 10–12: keep, revise or defer the wording, and explain why.
Sources and scope
Published by Lexifour. Software and AI assist with drafting and source review. Practical routines are editorial suggestions, not a guaranteed learning result.
- SKELL – examples and collocations for learners of English
- Official documentation verifies Examples and Word Sketch functions, automatic generation, the recorded cook/try combinations and synonym warning, and the need to identify versions when referencing results. Checked 2026-11-16.
- SkELL: Web Interface for English Language Learning
- The developers’ original 2014 paper documents historical sentence-level processing and example scoring; it is not used to assert the exact composition of the current corpus. Checked 2026-11-16.