TOEFL · Speaking · Listen and Repeat
Two or three meaning groups instead of a string of words — where to place the split, how to keep the ending, and what the scored attempts actually show.
By LingoLeap Research Team · Updated September 2026
Hear the whole line
Catch the action, object, and numbers
Compress to 2–3 chunks
Keep one complete idea in each
Reconnect the sentence
Chunk silently, speak continuously
Land the ending
Protect -s, -ed, numbers, and the last word
Chunking means holding a Listen and Repeat sentence as two or three meaning groups rather than a list of separate words. Working memory can only keep a handful of items at once, so grouping the words into a few chunks lets you carry a longer sentence without the end of it slipping away. Across 152,218 scored attempts, the share of a sentence a speaker reproduced fell from 78.8% on short sentences to 50.0% on the longest, and the highest-scoring deliveries used the fewest pauses. The practical rule that follows: break the sentence into two or three whole-idea groups, say it as one smooth line, and protect the ending — word-final sounds are dropped most.
Key findings
The length of the sentence — not the difficulty of its words — is what decides the score. The share a speaker reproduces falls from 78.8% at six words or fewer to 50.0% past sixteen.
The reason is how verbatim memory works. Holding a sentence word-for-word uses a short phonological store that fades in roughly two seconds; each extra word competes for the same few slots. Once the sentence needs more room than the store has, the overflow is lost — and it is lost from the end, because the end is what you reach last and rehearse least.
| Sentence length | Attempts | Avg score / 5 | Sentence retained |
|---|---|---|---|
| ≤6 words | 18,892 | 3.74 | 78.8% |
| 7–9 words | 47,733 | 3.51 | 73.7% |
| 10–12 words | 49,501 | 3.15 | 65.8% |
| 13–15 words | 23,034 | 2.79 | 57.2% |
| 16+ words | 13,058 | 2.61 | 50.0% |
Best practice
Never try to hold a long sentence word by word. As you listen, collapse it into two or three meaning groups — "the reading room closes at nine / on weekends / so plan accordingly" — and rehearse the last group first, since it is the most exposed.
A chunk is a group of words that arrives as one idea. It is not a grammatical unit and you do not need to name it. Compare:
As 13 words
The · reading · room · closes · at · nine · on · weekends · so · plan · your · visit · accordingly
As 3 chunks
The reading room closes at nine / on weekends / so plan your visit accordingly
Listen and Repeat is not a list of context-free sentences
Prompts usually come from campus life. Recognising the scene helps place the location, action, and time.
“The reading room closes at nine on weekends.”
“The shuttle leaves from the north entrance every fifteen minutes.”
“Please upload your assignment before class on Thursday.”
The data is clear that how many times you pause matters — one clean group beats a stop-start delivery (see below). Exactly where the break lands matters less than teachers often claim, so treat these split points as a way to keep each group a whole idea, not a rule to obey.
01Hear the sentence as groups, not words
Play a sentence and say aloud where it naturally falls into two parts. Do not think about grammar labels — think about which words arrive together as one idea.
02Learn the four split points
Break before a preposition (from, at, on, to), before a conjunction (and, but, so, because), after a conditional or time clause, and between an action and its details.
03Start short, then climb
Work at 6–9 words until you can repeat five in a row cleanly. Only then move to 10–12, then 13–15. Climbing early is the single most common way people stall.
04Hold chunks, deliver one sentence
Chunking is a memory device, not a delivery style. Rebuild the sentence and say it as one continuous unit with at most one or two brief pauses.
05Protect endings and numbers
Word-final -s, past-tense -ed, and any number or time is the first thing working memory drops. Say the ending deliberately in the last chunk.
06Bank fixed phrases as single units
"as soon as possible", "every fifteen minutes", "the front desk" — store these as one item so they cost one slot of memory instead of four.
Here is the whole method on one real 13-word sentence — the kind of length where scores start to fall. Watch where the chunks go, which words are traps, and what separates a 4 from a 2.
The sentence you hear
“The reading room closes at nine on weekends, so plan your visit accordingly.”
Break it into three chunks
Where this sentence trips people
A high-scoring attempt
Repeats all thirteen words with one short pause before "on" and one before "so", both plural endings intact, and "accordingly" landed clearly. Reads as one recalled sentence — a 4.
A low-scoring attempt
"The reading room close at nine on weekend, so plan your… visit." Both endings dropped, a mid-phrase pause after "your", and the final word lost. Same sentence, a 2 — and every difference is one of the four findings above.
Fewer pauses score higher. Near-continuous deliveries averaged 3.68 / 5; breaking the sentence five or more times fell to about 2.22 / 5.
Counting gaps longer than 250ms between words gives the number of chunks a speaker actually used. Each audible pause works against you twice: it tells the scorer the sentence was reassembled piece by piece rather than recalled whole, and it breaks the rhythm the prosody score rewards. Chunking is meant to happen in your head, not in your mouth.
| Pauses used | Attempts | Avg score / 5 |
|---|---|---|
| None (straight through) | 66,778 | 3.68 |
| 1 | 42,342 | 3.11 |
| 2 | 25,452 | 2.78 |
| 3 | 11,930 | 2.55 |
| 4 | 4,270 | 2.41 |
| 5+ | 1,446 | 2.22 |
Best practice
Hold two or three chunks mentally, then deliver the whole sentence as one line. If you must pause, pause once, at a natural boundary (before a preposition or conjunction) — not mid-phrase.
Word-final -s is one of the first things to go: across 237,757 words that carry it, the ending was dropped 11.4% of the time, and even when kept it scored only 74.3 / 100.
Endings fade first for two reasons at once. They arrive when memory is most loaded — at the end of the sentence — and sounds like -s and -ed are acoustically brief, so a rushed or uncertain delivery clips them entirely. The same is true of numbers and times, which carry meaning but little stress.
Ending protected
The room closes on weekends.
Ending lost
The room closes on weekends.
Best practice
Put the plural, the past tense, and any number or time into your last chunk, and rehearse that chunk first. Slow down very slightly on the final word rather than trailing off.
The single hardest sound is /θ/ — the "th" in thin — at 61.6 out of 100, mispronounced on 37.6% of attempts. That is far below every other sound measured.
Retaining the words is only half the task; the accuracy score also depends on how each sound is produced. Because the assessment scores pronunciation phoneme by phoneme, we can rank the sounds themselves — something a word-count study cannot. The reason /θ/ is so hard is that it has no equivalent in many first languages, so speakers substitute the nearest available sound, usually /s/ or /t/. The nasal ng and the vowel in goat come next.
| Sound | As in | Times produced | Avg accuracy | Below 60 |
|---|---|---|---|---|
| /θ/ | thin | 17,575 | 61.6 | 37.6% |
| /ŋ/ | sing | 61,203 | 69.8 | 27.2% |
| /z/ | zoo | 188,178 | 74.3 | 23.0% |
| /oʊ/ | goat | 83,991 | 75.2 | 25.0% |
| /s/ | see | 286,021 | 79.3 | 18.8% |
| /d/ | day | 212,537 | 80.7 | 16.5% |
| /ɝ/ | bird | 24,399 | 89.7 | 9.0% |
| /ʃ/ | shy | 56,794 | 90.8 | 7.4% |
| /ɔɪ/ | choice | 12,276 | 92.2 | 6.3% |
Best practice
Pronunciation gains come from a short, targeted list, not general practice. Spend five minutes a day on minimal pairs for your weakest sounds — thin vs. sin, thought vs. taught, think vs. sink — and re-record until the pair sounds distinct.
Words altered or dropped most often, out of those seen at least 30 times.
| Word | Times heard | Altered | Avg accuracy |
|---|---|---|---|
| turned | 61 | 100.0% | 41 |
| loaner | 104 | 100.0% | 44 |
| portraits | 189 | 99.5% | 54.2 |
| closed | 391 | 96.4% | 61.3 |
| sightlines | 314 | 100.0% | 62.7 |
| known | 149 | 98.7% | 63.3 |
| rollers | 263 | 100.0% | 65.5 |
| turnstiles | 315 | 97.8% | 67.2 |
| brochures | 74 | 98.7% | 68 |
| counters | 91 | 100.0% | 68.6 |
| valve | 90 | 96.7% | 69.9 |
| reentry | 799 | 99.6% | 72.7 |
The plan follows the one rule the data supports most strongly: master a length before you add to it. Each stage only moves up once the previous one is automatic, so you are never practising at the length where retention has already collapsed.
| Days | Sentence length | Target |
|---|---|---|
| 1–2 | 6–9 words | Five clean repeats in a row, one pause each |
| 3–4 | 10–12 words | Break before the preposition, keep every ending |
| 5–6 | 13–15 words | Three chunks, deliver as one sentence |
| 7 | Mixed | Random lengths, no warm-up |
Practise with the graded sentences, and work through the rest of the Listen and Repeat topic:
Each of these is common, and each has a specific fix.
Chunking out loud. Making the pauses audible turns a memory aid into hesitation — three breaks in a nine-word sentence reads as a stop-start delivery and costs both the fluency and prosody scores.
Fix: Keep the chunks in your head. Group the sentence silently, then say it as one continuous line.
Splitting inside a phrase. Breaking after "your" instead of before "so" cuts a whole idea in half, which is harder to hold and sounds unnatural.
Fix: Break only at boundaries — before a preposition or conjunction, after a time or condition clause.
Climbing too fast. Moving to 13-word sentences before 9-word ones are automatic is the most common plateau; retention has already started to fall off in that band.
Fix: Stay at each length until you can repeat five in a row cleanly before adding words.
Swallowing the endings. Word-final -s is dropped on about one in ten of the words that carry it, and it is the difference between "close" and "closes".
Fix: Put endings in your last chunk and land the final word deliberately instead of trailing off.
Translating while listening. Any translation step spends the very working-memory capacity that chunking is meant to free up.
Fix: Repeat by sound and rhythm, not meaning — you do not need to understand a sentence to reproduce it.
Record one sentence and get your own version of the tables above — which words you dropped, ending survival, and where your pauses landed.
Try a free Listen and Repeat setWhat was analyzed. 152,218 scored Listen and Repeat items from 23,390 real practice sessions on LingoLeap — not a survey or a lab study, but actual attempts. Timing and accuracy are measured for every word by automatic pronunciation assessment, down to the individual phoneme; each sentence also carries a LingoLeap AI score on the 0–5 per-item scale.
How the figures are defined. A pause is a gap longer than 250ms between two words. "Retained" is the share of the target sentence a speaker reproduced, recovered by aligning the target to the recognised words. Sentence-length and phoneme figures are averaged over every attempt in the window.
Limits, stated plainly. AI scores are not official ETS scores. Because omissions are recovered by alignment, a word marked dropped may occasionally have been said unclearly rather than skipped. Users self-select into practice, so this is a picture of engaged learners, not a random sample of test takers — read the directions as reliable and the exact magnitudes as indicative.
Chunking means holding a sentence as two or three meaning groups instead of a string of individual words. Working memory holds a small number of items, so grouping words lets you carry a longer sentence without losing the end of it.
Two or three. Score falls with every extra pause in the data, and a stop-start delivery reads as hesitation — so hold two or three groups in your head, but say the sentence as one smooth line.
Before a preposition or conjunction, after a conditional or time clause, or between an action and its details. In our data, exactly where the break lands matters less than how few pauses you make — so use split points to keep each group whole, not as a strict rule.
Because you run out of room to hold them. The share of a sentence a speaker reproduces drops steadily as it gets longer, and the score drops with it. Chunking lets you carry more of the sentence at once.
About a week of short daily sessions is enough to change how you hold a sentence. Accuracy on 13–15 word sentences takes longer, since that band is where scores fall most sharply.
Partly. The memory benefit is specific to repetition, but the habit of grouping words into phrases improves rhythm and reduces mid-phrase hesitation in Interview answers as well.
No — and trying to can hurt. Repetition is scored on sound, fluency, and completeness, not comprehension. Any effort to translate or parse the meaning spends the working-memory capacity you need to hold the words, so repeat by sound and rhythm and let meaning look after itself.
Keep going and reconstruct the shape of the sentence rather than stopping to search for the exact word. Completeness rewards reproducing most of the sentence smoothly; a long mid-sentence pause to recover one word usually costs more than the missing word itself.
Start with /θ/ (the "th" in thin), which scored lowest by a wide margin, then the nasal "ng" and the vowel in "goat". Five minutes a day of minimal pairs — thin vs. sin, thought vs. taught — raises the pronunciation accuracy score faster than general repetition.
Each sentence is scored 0–5 on how accurately and completely you reproduce it, combining pronunciation accuracy, fluency, and how much of the sentence survives. A seven-item set gives a raw total out of 35, which maps onto the section score.
Listen and Repeat: format and scoring
What the task is, timing, and how the 0–5 per-sentence score works.
Seven Listen and Repeat strategies
Shadowing, rhythm, and memory tactics beyond chunking.
Campus scenarios and practice sentences
Thirty graded sentences across eight scenario types.
Sentence breakdown: the conference schedule
One sentence, 977 real attempts, and the words most often dropped.