An experiment that didn’t stop at warning us about noise — and the third study that found the way through
Introduction: There Is No Quiet House
A house with a child in it cannot be quiet. The TV is on in the living room, the kitchen fan is running, a sibling is chattering nearby, someone is on the phone. And into all of it, we speak to the child. “What’s this? This is a cup.”
In that ordinary scene, can the child learn the word cup?
We hear plenty about background noise being bad for children. But most of it stops there. The advice to reduce noise is not realistic. No household can run three meals a day at library volume.
McMillan and Saffran, at the University of Wisconsin–Madison’s Waisman Center, dug into the problem differently. They did not stop at confirming that noise interferes. They went on to ask what, then, protects the child. The answer is in their third experiment.
The Core Question
The team’s question comes in three stages.
“When people are talking in the background, can a child learn a new word?”
- How loud does it have to be before learning collapses?
- As a child grows and their vocabulary expands, do they become able to overcome noise?
- Is there a condition that lets them learn even inside the noise?
How They Studied It: Teach in Noise, Test in Quiet
To understand the design you need to know its three stages.
Stage 1. Getting Used to the Sound (about 1 min 15 sec)
The child hears sentences containing invented words. coro, tursey — words that do not exist. No objects are shown yet. Only the sound.
Stage 2. Linking Sound to Object (about 2 min 30 sec)
Now an unfamiliar object appears while the word plays. This is the real learning stage, pairing sound with thing.
Stage 3. Testing
Two objects appear on screen and the child hears something like “Where’s the coro?” If the child looks longer at the correct object, they are counted as having learned the word. Gaze was coded frame by frame in 33-millisecond units.
Here is the decisive part of the design. The noise was present only in Stages 1 and 2; the Stage 3 test was always quiet. So this experiment is not measuring “they couldn’t make it out because it was loud.” It measures whether noise intruding at the moment of learning broke the learning itself.
What the Noise Was
The background was not TV sound or machine hum but human voices. Sentences read by native male English speakers (the Harvard IEEE corpus) were randomly overlaid to create the sound of two people talking at once. Closer to adults conversing in the next room.
Volume was set like this.
| Condition | Speech to the child | Background voices | Difference |
|---|---|---|---|
| Easy condition | 65 dB | 55 dB | 10 dB |
| Hard condition | 65 dB | 60 dB | 5 dB |
A 10 dB gap means the target voice is distinctly louder than the background. A 5 dB gap means only slightly louder — both audible, with no obvious answer to which one you should attend to.
Design and Participants
- Experiment 1: 40 children aged 22–24 months (mean 23 months). 20 in the 10 dB condition, 20 in the 5 dB condition
- Experiment 2: 40 children aged 28–30 months (mean 29.1 months). Again 20 per condition
- Experiment 3: 26 children aged 28–30 months (mean 28.3 months)
- All from monolingual English environments
- Trials: 24 in Experiments 1 and 2, 18 in Experiment 3
- Gaze measurement windows: before the word appeared (baseline) and from 0.3 seconds after the word appeared (target window). The 0.3-second offset subtracts the time it takes to move the eyes
Result 1: At 5 dB, Learning Stopped
The key number is the proportion of looking to the target. How far the child’s gaze shifted toward the correct object after hearing the word, compared with before. If they learned, it should rise.
22–24 months (Experiment 1)
- 10 dB condition: .53 → .66 (p < .001, d = .89) — they learned
- 5 dB condition: .55 → .55 (p = .960, d = .01) — nothing moved at all
An effect size of 0.01 in the 5 dB condition is essentially zero. Not learned a little less — learned nothing.
28–30 months (Experiment 2)
- 10 dB condition: .53 → .67 (p < .001, d = .95) — they learned
- 5 dB condition: .57 → .64 (p = .106, d = .38) — not statistically significant
This one has to be read honestly. The older children’s 5 dB condition was not a flat zero; there was a slight upward tendency that failed to clear the statistical bar. That is different in kind from the younger children’s total non-response. Still, it was not at a level where you can say they learned.
Neither Six Months Nor a Bigger Vocabulary Helped
This part is the surprise. The children in Experiment 2 were six months older than those in Experiment 1 and had far more expressive vocabulary (mean 77.5 words vs. 56.75, p < .001). And yet the pattern of vulnerability to noise was identical across the two groups.
Getting better at talking did not bring with it the ability to learn words in a noisy environment.
Result 2: The Turn in the Third Experiment
Had it ended there, this study would have closed on the familiar conclusion: reduce the noise. The team went one step further.
The Design
They taught four new words to 26 children of the same 28–30 month range. But in two kinds.
- Two pre-heard words: in Stage 1 (getting used to the sound) these were heard first in quiet. No object yet, just the sound cleanly
- Two novel words: no prior exposure
And Stage 2 learning ran in 5 dB noise for all four words — precisely the condition where learning had collapsed in the earlier experiments. Testing was again in quiet.
Because the two kinds were compared within the same child, differences in individual ability are controlled.
The Result
- Pre-heard words: .52 → .66 — learned
- Novel words: .53 → .55 — not learned
The statistics split too. Main effect of prior exposure F(1,25) = 9.21, p = .006, partial eta squared .27. Interaction of prior exposure and gaze window F(1,25) = 5.02, p = .034.
The same child, in the same noise, for the same amount of time — and the outcome split on whether they had heard that sound once in quiet beforehand.
Why Hearing It First Protected Them
The team’s explanation:
Experience with clean, fluent speech may support the ability to perceive and acquire words in subsequent difficult listening conditions.
In everyday terms. When a child meets an unfamiliar word in a noisy room, they have to do two things at once. Pull that chunk of sound out of the noise, and attach that sound to the object in front of them. For a toddler this dual task is more than they can carry.
But if they have already heard that sound in quiet, the first job is largely finished. The contour of the sound is already in their head, so they can recognize it even in noise and spend what is left on linking it to the object.
What the experiment shows, then, is that removing noise is not the only answer. Giving the child a chance to become familiar with the sound first resolved a good deal of it.
Practical Takeaways
Plant New Words in the Quiet Moments
This is the study’s most concrete instruction. Words you are teaching a child for the first time are better let loose during low-noise times. Before sleep, in the bath, alone together in the car — any moment when the background is empty.
What matters is that you do not have to teach it perfectly on the spot. What the children in Experiment 3 did in the quiet was simply hear the sound. They saw no object and knew no meaning. It lasted 1 minute 15 seconds. That alone brought later learning in a noisy room back to life.
Making Moments Is More Realistic Than Making the House Quiet
An all-day quiet house is impossible. This study says you don’t need one. What is needed is not a reduction in total noise but an empty moment at the point where a new word first arrives.
Put the other way around: telling a child the name of an unfamiliar object while the TV is on and several people are talking may be a low-efficiency attempt.
This Matters More Where Quiet Is Hard to Come By
The team raises this in the discussion. Children from lower-income households are exposed to noisier environments more often. Homes are more crowded, TV viewing hours are longer, the schools they attend are louder. The team concludes that household background noise may be one more factor shaping a child’s language learning.
Which is where the third experiment’s value grows. A home’s noise level is hard for an individual to change; making a short quiet moment is comparatively possible.
Limitations
- The background was two voices only. The team names this themselves. Whether the same result holds with a different number of talkers, or with other kinds of noise like TV, music, or machines, is unconfirmed.
- The paper does not say what 5 dB or 10 dB looks like in your actual house. Signal-to-noise ratio is the relative difference between two sounds, not an absolute noise level. Translating it straight into “don’t leave the TV on” is an interpretation the study does not warrant.
- The taught words were too different from one another. tursey, coro, pif, blicket are clearly distinguishable by sound. The team acknowledges this may have made the task easier than reality. With similar-sounding words the results might have been worse.
- Many children were excluded. 24 in Experiment 1, 21 in Experiment 2, 12 in Experiment 3 dropped from analysis. Mostly children who could not stay focused enough to reach the test stage or complete half the trials. Common in infant eye-gaze research, but it also means only the children who sat through to the end remain.
- It was a lab setting, and short. Under four minutes of learning in total. Different from real language acquisition, repeated at home over months.
Closing Thoughts
The sentence this study leaves is not a warning.
Noise blocked the children’s learning, but the children who had heard the sound beforehand learned inside the same noise.
A demand to make the house quiet is unworkable in most homes. This study asks for something smaller. At the moment a new word first arrives for the child — leave that one moment empty.
Children are more vulnerable to a loud world than we think. And they withstand it with less preparation than we think.
Source: McMillan, B. T. M., & Saffran, J. R. (2016). Learning in complex environments: The effects of background speech on early word learning. Child Development, 87(6), 1841–1855. https://doi.org/10.1111/cdev.12559