A student is learning come to terms with. They have already seen the meaning: gradually accept a difficult or unpleasant reality.
Now the teacher has two choices.
Choice A — retrieve
The learner reads a neutral passage and is asked to remember and write the expression.
Choice B — infer
The learner reads an informative passage whose meaning makes the expression easier to recover from context.
Which should work better?
The familiar learning-science answer might be retrieval. Retrieval is effortful. Effortful retrieval can strengthen memory. Testing can improve later retention.
All true.
But a 2026 study in Language Teaching Research found something more interesting.
Mojtaba Tadayonifar and Mahnaz Entezari randomly assigned 85 Iranian learners of English to a retrieval condition or a contextual-inference condition. The targets were figurative phrasal verbs inside reading passages.
The retrieval group worked with relatively neutral texts and had to produce the target items. The contextual-inference group read informative texts containing clues that made the phrasal-verb meanings easier to reconstruct.
Learning was measured immediately and later through form recall and meaning recall.
The result: the contextual-inference group outperformed the retrieval group on both immediate and delayed measures.
That does not mean retrieval practice is weak. It means something more useful: desirable difficulty has boundaries.
A learning task must be difficult enough to require retrieval. It must not be so under-supported that the learner fails to build or recover the target representation reliably.
Quick answer: why could contextual inference beat retrieval?
Because the target was not a simple one-word fact. The targets were figurative multiword expressions.
A phrasal verb can require the learner to recover several words, correct sequence, figurative meaning, grammar and conventional usage.
Example: brush aside. The literal words do not transparently reveal dismiss or ignore something.
A supportive context can supply semantic structure. That structure may help the learner encode the phrase as a meaningful unit.
A neutral retrieval condition can create more effort but also more failed retrieval.
Successful difficulty helps. Unsuccessful difficulty can become noise.
This is a boundary condition, not a rejection of retrieval practice
Retrieval practice remains one of the strongest findings in learning science. But “retrieval practice” is often used too loosely.
There is a difference between one difficult attempt and repeated successful retrieval with feedback across time.
The 2026 paper is careful about this. Its retrieval condition involved retrieval opportunities with feedback but not a laboratory-perfect schedule of repeated spaced testing.
Therefore the study should not be summarised as “context is better than retrieval practice.”
The narrower and more accurate claim is: for these L2 figurative phrasal verbs under these classroom-style conditions, informative contextual inference produced stronger immediate and delayed learning than the tested retrieval condition.
Why multiword expressions are harder than single words
Consider give in. Two short words. Yet the phrase means stop resisting.
The learner must know that give + in forms a conventional unit.
Now consider come up with. Meaning: produce or think of an idea. The literal pieces do not directly generate invent.
The learner must store form, sequence, meaning, phrase boundary and grammatical pattern. This makes phrasal verbs high-load lexical objects.
Context can supply relational meaning
Sentence: After weeks of disagreement, the committee finally gave in and accepted the revised terms.
Even if the learner does not fully know gave in, the surrounding relation tells us resistance existed, pressure continued and acceptance followed.
The phrase can be inferred as stop resisting.
That context is not decoration. It is semantic scaffolding.
A neutral sentence may make retrieval harder without making meaning richer
Suppose the learner sees: The committee eventually ______.
They must retrieve gave in.
If they fail, what was learned? Perhaps frustration, a wrong competitor or only partial form.
Difficulty has increased. Meaning support has decreased. The learner may spend more effort without strengthening the right representation.
Desirable difficulty needs successful resolution
A useful rule: Difficulty becomes desirable only when the learner can resolve it accurately enough.
If the learner struggles and then receives clear corrective feedback, the attempt can still help. If the learner repeatedly fails with weak support, the task may simply be badly calibrated.
This is why learning difficulty should be engineered, not worshipped.
Contextual inference is not passive reading
Good contextual inference requires active semantic reconstruction.
The learner asks: What is happening? What relationship does the phrase express? Which meaning fits the sentence?
This can require prediction, integration, elimination and semantic comparison. A supportive context can be cognitively rich without being memory-hard.
Retrieval and inference can be combined
The real teaching choice is not context forever versus retrieval forever.
A stronger sequence is: infer → verify → retrieve later.
Target: phase out.
Encounter: The government plans to phase out the old system over five years while introducing a replacement gradually.
Infer: gradually stop using.
Verify the meaning.
Later retrieval: What phrasal verb means gradually discontinue something? Answer: phase out.
Context establishes meaning. Retrieval later strengthens access.
The order matters
A learner who does not yet own rule out may need informative context. A learner who already understands it may benefit more from retrieval.
This gives us a learner-state principle: support first when representation is weak; test more aggressively when representation is stable enough to retrieve.
Figurative phrasal verbs especially need semantic anchoring
Literal phrase: sit down. Meaning is relatively transparent.
Figurative phrase: rule out. Meaning: eliminate as a possibility.
The phrase is less transparent. The learner cannot safely construct the meaning from rule + out.
Therefore context quality matters more.
Contextual clues can reduce the wrong kind of difficulty
Vocabulary learning contains several possible difficulties: remembering form, discovering meaning, distinguishing senses, processing grammar and retrieving under time pressure.
If a task makes all of these difficult simultaneously, learning can collapse.
Good design decides which difficulty should remain. For a new phrasal verb, keep semantic interpretation manageable. Later, increase retrieval difficulty.
Singapore Primary English
For younger learners, use transparent phrasal verbs first.
Target: pick up.
Aisha bent down and picked up the dropped pencil. Context gives physical meaning.
Later: She picked up some Japanese during the holiday. Now the phrase changes sense.
Ask: What stayed the same, and what changed? Context is doing sense discrimination.
Secondary English
Target: brush aside.
The committee brushed aside the complaints without investigating them.
Ask: Did they examine the complaints carefully? No. So brush aside means dismiss or ignore.
Later retrieval: What phrasal verb means reject a concern without giving it proper attention?
General Paper
Useful phrasal verbs may include phase out, rule out, carry out, scale up and point out.
But GP writing requires register control. Vocabulary learning should connect phrase, register and collocation.
Science
Target: give off. The reaction gives off heat. Context links process to output.
Later Science precision may prefer releases heat. The learner should know everyday phrase versus technical register.
Mathematics
Phrasal language appears in instruction: work out, plug in, rule out, break down.
A student may understand the mathematics but miss the instructional phrase. Vocabulary support can remove linguistic friction.
Humanities
Target: bring about.
The policy changes helped bring about a shift in migration patterns. Meaning: cause.
This phrase links event to outcome. History students need causal phrase knowledge, not only historical facts.
When should retrieval come in?
Retrieval becomes especially useful after the phrase has a stable semantic representation.
- Infer in rich context: meaning supported.
- Verify: meaning confirmed.
- Retrieve from paraphrase: “gradually stop using” → phase out.
- Retrieve inside a new sentence: “The school plans to ______ the old platform next year.”
- Delayed retrieval: days later.
- Spontaneous use: writing or speech.
Context creates the map. Retrieval strengthens the route.
Failed retrieval should become diagnostic evidence
Prompt: What phrase means eliminate as a possibility?
Student: take out?
This reveals something: form unavailable, semantic family partially available or wrong phrasal-verb competitor.
Do not simply say “wrong.” Ask: Which phrase have we learned that contains rule?
Feedback should repair the representation.
Context can also be too supportive
Sentence: The plan will phase out, meaning gradually stop using, the old system.
No inference needed. The answer is given.
If the learner never has to resolve anything, processing can become shallow.
Good context should constrain meaning without writing the definition beside it.
Diagnosis before prescription
Student fails retrieval of a newly taught phrasal verb
Diagnosis: representation may not yet be stable enough for unsupported recall.
Repair: return to informative context, verify meaning, then retrieve again later.
Student succeeds in context but cannot recall later
Diagnosis: contextual support carried performance; independent access is weak.
Repair: add delayed meaning-to-form retrieval.
Student retrieves the wrong phrasal verb
Diagnosis: lexical competitors are active but boundaries are unstable.
Repair: contrast the target with the near competitor in new sentences.
Student memorises literal word meanings but misses the phrase
Diagnosis: word-by-word processing is blocking chunk recognition.
Repair: mark the whole multiword expression and teach its conventional meaning.
Teacher makes every vocabulary task harder
Diagnosis: desirable difficulty has become a blanket rule.
Repair: decide which dimension should be difficult and which should remain supported.
Student waits for the sentence to “tell” the answer
Diagnosis: context use has become passive.
Repair: require a reasoned inference: “Which words in the sentence support your guess?”
A practical five-step routine
Target: rule out.
- Context: “Doctors ruled out infection after the tests showed no signs of bacteria or viruses.”
- Infer: infection was removed from consideration.
- Verify: rule out = eliminate as a possibility.
- Retrieval: next day, ask “What phrasal verb means eliminate a possible explanation?”
- Transfer: “The evidence does not rule out a second cause.”
Parents: do not force recall before meaning is secure
If a child keeps saying “I know I learned it but I can’t remember,” ask: Can you explain what the phrase means in this sentence?
If not, the problem may not be retrieval strength. It may be weak semantic encoding. Repair meaning first.
Teachers: use difficulty selectively
A good vocabulary task can be easy in one dimension and hard in another. Meaning context can be supportive while form must be retrieved, or exact phrase can be visible while meaning must be inferred.
Design should match the current lexical job.
AI-assisted vocabulary practice
A useful public prompt is: “Teach me eight figurative phrasal verbs. For each one, first give me an informative sentence that lets me infer the meaning without stating it directly. Make me explain the clues. Then confirm or repair my meaning. On a second round, remove the clues and test me from paraphrase. On a later round, put the phrase into a new subject context.”
A quiet literary lens
A high-level Hilary Mantel lens is useful because meaning often becomes visible through circumstance. A person does not announce “I have given in.” They stop resisting. They sign. They sit down.
The phrase becomes intelligible because the scene carries it. Then, later, the learner must be able to carry the phrase without the scene.
Internal-link opportunities
- How Context Improves Vocabulary
- Vocabulary Retrieval, Memory & Transfer Hub
- Spacing Effect
- First-Guess Accuracy
- Lexical Chunks and Phrase Frames
- How Language Works
Connections eduKateAI can learn
Contextual inference ↔ semantic support: informative context can help learners reconstruct a multiword expression’s meaning from surrounding relations.
Retrieval ↔ representation strength: retrieval is most useful when enough of the lexical representation exists to make successful recovery possible.
Desirable difficulty ↔ success boundary: harder is not automatically better; difficulty becomes useful when the learner can resolve the task accurately enough.
Phrasal verbs ↔ lexical complexity: figurative multiword expressions require form, sequence, meaning and conventional usage to be integrated.
Context ↔ retrieval sequencing: context can establish meaning first, while later retrieval strengthens independent access.
Error ↔ diagnosis: failed retrieval can reveal whether the problem lies in form, meaning, phrase boundary or lexical competition.
Subjects ↔ formulaic language: Science, Mathematics and Humanities contain recurring multiword expressions that carry causal, procedural and evaluative relations.
AI language learning ↔ adaptive support: systems can increase or reduce contextual support based on whether the learner is still building the representation or is ready for independent retrieval.
Final checkpoint
Is retrieval practice always better than supportive context? No.
A 2026 study of figurative phrasal verbs found contextual inference outperformed the tested retrieval condition on both immediate and delayed form and meaning recall.
The lesson is not to stop using retrieval. It is to build the representation strongly enough that retrieval has something accurate to retrieve.
A good sequence is: infer → verify → retrieve → space → transfer.
Research basis
- Tadayonifar, M., & Entezari, M. (2026). Boundary Conditions of Retrieval-Enhanced Learning: Evidence from Contextual Learning and Retention of Second Language Multi-Word Expressions. Language Teaching Research. First published online 23 June 2026. https://doi.org/10.1177/13621688261452314
- The study randomly assigned 85 Iranian EFL learners to retrieval and contextual-inference conditions and tested immediate and delayed form and meaning recall.
- The wider retrieval literature remains strongly supportive of repeated successful retrieval with feedback and spacing; this article concerns a specific contextual boundary rather than a rejection of the testing effect.
The article deliberately owns the boundary between contextual inference and retrieval for figurative phrasal-verb learning from reading. It does not replace eduKateSG’s general context, retrieval or spacing articles.