Listen to a sentence five times and it becomes familiar. Cover the text in a real situation, however, and you may still think, "I know this, but I cannot say it."
The missing step is often not more exposure. It is retrieval: producing language from a cue instead of recognising it on the screen. Retrieval practice means attempting the expression first, then checking a model, correcting it and retrieving it again.
What the vocabulary study actually found
In a 2013 study, Kang, Gollan and Pashler compared imitation with retrieval practice for foreign spoken vocabulary. One group heard a word and repeated it; the other saw a picture, tried to produce the word and then heard it. On immediate and two-day-delayed tests, the retrieval group showed better comprehension and production of the words, with no loss in pronunciation quality.
That is evidence about foreign vocabulary learning, not proof that any unscripted speech will improve through recall alone. Its practical lesson is narrower and useful: the model does not always need to come before your attempt. Trying first reveals what you can and cannot retrieve.
Extending this principle to short functional sentences, as below, is a practice recommendation rather than a direct result tested by that experiment.
A five-step cycle
1. Keep a cue, hide the answer. Write "ask a colleague to reply by the end of today", choose a picture, or leave three keywords. Do not display the complete English sentence. 2. Say a sentence that completes the task. Simple and hesitant is acceptable. If one word will not come, describe around it rather than translating silently forever. 3. Check or hear a reliable model. Compare how each version accomplishes the task. A different wording is not automatically wrong. 4. Repair, then retrieve again. Say the corrected version once with support, then once using only the original cue. 5. Change the cue later. Move from asking a colleague for a reply to asking a landlord when the repair will happen. The function stays; the context changes.
Three useful sentences completed through the whole cycle reveal more than twenty answers read silently.
Choose cues that represent a communicative need
An intention in your first language is practical. For "that is not what I meant", try first, then compare *That's not quite what I mean.* You are practising access from meaning to English, not recalling a sentence from its first letter.
A picture or situation reduces word-by-word translation. See hair in a restaurant soup and explain the problem to the server. You need a complete message, not just the word *hair*.
Keywords support slightly longer output. With *deadline / Friday / confirm*, you might produce *Could you confirm by Friday whether you can meet the deadline?* Start with one sentence; unscripted speeches are not the entry requirement.
Retrieval, shadowing and spacing do different jobs
Retrieval practice controls the order: attempt before the answer. Shadowing supplies a sound and rhythm model: listen and reproduce. Spaced review controls when you return: tomorrow, three days later, next week.
Combine them: speak from a cue, compare a natural model, shadow its audio, hide the text and say it again, then retrieve it in a new situation tomorrow. They are not competing miracle methods.
For return dates, read Spaced Repetition for Speaking. For working with the sound model, see How to Practise English Speaking Alone.
Four common mistakes
Treating a stall as failure. A stall tells you where feedback is needed. Judge the cycle by whether you can retrieve a better version afterwards.
Guessing repeatedly without feedback. If you do not know, stop guessing and find a trustworthy model. Rehearsing an incorrect version fluently is not the objective.
Using a grammar chapter as the cue. "Politely request an extension" represents a task. "Review unit three on conditionals" represents a syllabus.
Keeping the cue identical forever. Change the person, time and place. Real communication rarely presents the same flashcard twice.
MirrorSay belongs after the first attempt
Give MirrorSay your own sentence based on an intended meaning. Compare the natural formulation, hear and shadow the model, then hide the text and speak again. The order matters: your first attempt gives the feedback something to work on.
When just one word is missing, keep the message moving with Paraphrase When You Forget a Word.
References
- Kang, S. H. K., Gollan, T. H., & Pashler, H. (2013). Don't just repeat after me: Retrieval practice is better than imitation for foreign vocabulary learning. *Psychonomic Bulletin & Review*, 20, 1259–1265. Article and abstract
*MirrorSay is a free AI speaking-practice app: say a sentence, see a natural way to express it, then shadow the model. No sign-up, and speech is processed on-device whenever possible.*