Language · Neuroscience · Psychology
The language a child hears
The neuroscience of conversation, the psychology of belonging, and what the parent-talk research line has measured.
A child understands a grandparent’s story well enough to ask the next question. That ordinary exchange is what the research below tries to explain: the sounds, the relationships, and the support that make it possible.
How does a language become part of a relationship?
- Caregiver voice
- Child attention
- Child reply
- Adult response
A conceptual conversation loop. The arrows describe an exchange, not a measured brain pathway.
Live sound and infant discrimination
In two experiments, Kuhl, Tsao, and Liu studied English-learning infants around nine months old. The first compared live Mandarin exposure with live English; the second tested audiovisual and audio-only Mandarin recordings. Each involved twelve sessions.
Live Mandarin exposure supported discrimination of a Mandarin speech contrast. The recorded conditions did not show the same learning. The comparison makes the social setting part of the question.[1]
- infant age
- 9 mo
- sessions each
- 12
- experiments
- 2
Exposure condition
Live interaction
Experiment 1: infants learned the Mandarin sound contrast after live exposure, compared with English exposure.
Qualitative schematic. Arrows distinguish responsive interaction from recorded exposure.
Limitation
The outcome was infant sound discrimination. It cannot establish later fluency or settle how children of other ages learn from screens.
Conversational turns and the listening brain
Romeo and colleagues paired home recordings with a story-listening brain-imaging task in 36 children, ages 4 years 6 months to 6 years 10 months.
More adult–child conversational turns were associated with greater activity in the left inferior frontal region, including Broca’s area. The association remained after accounting for socioeconomic status and adult word counts. The study links everyday interaction to language-related brain function; it does not establish cause and effect.[2]
- children
- 36
- years old
- 4½ to 6¾
- story task
- fMRI
Schematic location only. The yellow region represents neither a scan nor an effect size.
Limitation
This small observational study cannot show that additional conversation caused the brain-activity difference. Brain activity is not an intelligence score.
Heritage ties across three generations
A language can hold an invitation to family life: a story understood, a joke answered, a conversation joined without translation.
In an August 2026 study, Priya Mariana Shimpi interviewed 16 parents from 12 Bay Area households considering Spanish or Mandarin education. For parents with heritage ties and limited proficiency, motivation coexisted with uncertainty, regret, and a desire to learn alongside their children.[3]
- parents
- 16
- households
- 12
- languages
- 2
Reading the diagram
The diagram sketches one possible family, rather than a universal sequence. Words, stories, and shared occasions offer different ways to participate. A parent can be a learner too. That possibility makes encouragement, low pressure, and room for mistakes relevant design questions.
Illustrative family, not the study’s data or its categories. Fluency and household arrangements differ across families.
Limitation
These are themes from a small selected sample, not estimates of how all families feel or evidence of a benefit to children.
Parent support, belief, and transfer
John List, Julie Pernaudet, and Dana Suskind tested two parent-support programs. Four short videos changed beliefs without lasting gains in parental investment or child outcomes. A separate twelve-visit home program improved interaction and children’s vocabulary, math, and social-emotional skills.[4]
The trials involved different ages and populations. They show why a program needs to measure what follows a lesson; they do not isolate the effect of extra coaching or time.
- short videos
- 4
- home visits
- 12
- trials
- 2
A proposed sequence for evaluating support. Arrows organize the questions; they do not identify each step as a proven causal mechanism.
AI as a research instrument
A June 2026 working paper coauthored by List and Suskind analyzes over 600 hours of parent–child speech from two home-visiting experiments. AI helps measure acoustic features in the recordings. The tested intervention supports parents; the paper does not evaluate an AI tutor. This summary draws on the official abstract.[5]
The research line behind the trials
The 2021 trials sit inside a longer program of work. Betty Hart and Todd Risley’s 1995 home observations, published as Meaningful Differences in the Everyday Experience of Young American Children, gave the field its name, the thirty-million-word gap, and a long argument over how the gap was measured.
Dana Suskind’s Thirty Million Words initiative at the University of Chicago, now the TMW Center for Early Learning + Public Health, turned that observation into interventions. A 2016 pilot trial tested an eight-week home-visiting curriculum with 23 mother and child pairs, the children aged one and a half to three. Parents’ knowledge of language development rose and was still higher four months later. Parent talk and child word types rose one week after the program ended; at four months those gains were no longer significant.[6]
The 2021 trials scaled the approach in two directions: short videos for several hundred newborn families, and twelve home visits for a smaller toddler group. The 2026 working paper returns to the recordings those visits produced and uses AI to measure acoustic and linguistic features of parental speech across more than 600 hours of audio.
An open question
The same pattern recurs across the line: knowledge and behavior change while a program is running, and the question is what persists afterward and what reaches the child. A finished lesson, an unprompted reply, and a stronger sense of connection are different outcomes. Each needs its own evidence. The explorer below turns that distinction into the questions a study would have to answer.
- 1995Hart and Risley publish the home observations behind the thirty-million-word gap.
- 2016Suskind’s eight-week home-visiting pilot: 23 mother and child pairs.
- 2021List, Pernaudet and Suskind: short videos for newborn families, twelve visits for toddlers.
- 2026Pernaudet and List: AI measures 600+ hours of recorded parent speech.
Proposed outcome
Program completion
- What it would measure
- Whether families take part in a program and finish its activities.
- What a test would require
- Clearly defined completion and return rates, with reasons for stopping. Participation must stay optional.
- What it would not establish
- Completion alone cannot show language learning or a change in family relationships.
No results are reported here. This control describes study designs that would be needed to answer each version of the question.
Limitation
These trials evaluate specific parent-support programs with specific families. They do not establish that any other program, curriculum, or product for children’s language works, and none of them measured heritage-language outcomes.
Sources and evidence
Six studies, with different methods and populations. The diagrams are editorial explanations; no plotted values or brain images are reconstructed.
- studies
- 6
- published
- 2003 to 2026
- randomized trials
- 3
- 1
2003 · Two infant experiments
Foreign-language experience in infancy: Effects of short-term exposure and social interaction on phonetic learningPatricia K. Kuhl, Feng-Ming Tsao & Huei-Mei Liu · PNAS
English-learning infants around nine months old. Live Mandarin vs. English in one experiment; audiovisual vs. audio-only Mandarin in another. Short-term sound discrimination, not fluency.
- 2
2018 · Observational neuroimaging
Beyond the 30-Million-Word Gap: Children’s Conversational Exposure Is Associated With Language-Related Brain FunctionRachel R. Romeo and colleagues · Psychological Science
36 English-speaking children, ages 4 years 6 months to 6 years 10 months. Home recordings and a story-listening fMRI task. Covariate adjustment does not establish causality.
- 3
2026 · Exploratory mixed methods
Choosing additive bilingualism: a language policy paradox for liminal heritage parentsPriya Mariana Shimpi · Frontiers in Communication · 19 August
16 parents from 12 Bay Area households considering Spanish or Mandarin education. Eight with heritage ties, eight without. Themes from a small selected sample, not population estimates.
- 4
2021 · Two randomized trials
Shifting parental beliefs about child development to foster parental investments and improve school readiness outcomesJohn A. List, Julie Pernaudet & Dana L. Suskind · Nature Communications
475 newborn dyads and 91 toddler dyads in separate studies. Intervention effects are evaluated within each trial. The comparison does not isolate program intensity.
- 5
2026 · Working paper · abstract reviewed
Leveraging Artificial Intelligence and Field Experiments to Explore Novel Features of Parental Speech and Foster Child DevelopmentJulie Pernaudet, John A. List and colleagues, including Dana Suskind · NBER Working Paper 35302 · 1 June
The official TMW summary describes AI analysis of more than 600 hours of parent–child speech from two home-visiting experiments. This page does not appraise the full working paper.
- 6
2016 · Randomized controlled pilot
A parent-directed language intervention for children of low socioeconomic status: a randomized controlled pilot studyDana L. Suskind, Kristin R. Leffel and colleagues · Journal of Child Language
23 mother–child dyads, children aged 1;5 to 3;0, eight weekly hour-long home visits. Parent knowledge gains held at four months; parent-talk and child word-type gains were significant at one week but not at four months.
The evaluation questions on this page are editorial suggestions, not study results. Hart and Risley’s 1995 book is named for its place in the research line; no figures from it are reproduced.
Related research dissection: Missing Women and the Price of Tea in China →