About

Elika Bergelson is an Associate Professor at Harvard University's Psychology Department. Her research focuses on early lexical development in infants and toddlers, exploring how visual and linguistic experiences interact in early word-learning. She investigates how infants learn language from the world around them, with a particular interest in the non-linear improvement in word comprehension and word learning known as the 'Comprehension Boost'.

Research topics

• Computer Science
• Psychology
• Developmental psychology
• Natural Language Processing
• Statistics
• Data science
• World Wide Web
• Mathematics education
• Cognitive psychology
• Cognitive science
• Neuroscience
• Social psychology

Selected publications

• Response to Creel

  2026-04-18

  articleOpen accessSenior author

  Our original paper (Liu, Hilton, et al., 2023) found effects of linguistic experience on music perception which generalized across a half-million speakers of fifty-four languages. In a critique, Creel (Creel, 2026) claims that our multilevel modelling approach biased the language-level estimates such that languages with small-sample sizes “spuriously appear to pattern consistently” compared to “raw” sample means. We respectfully disagree and counter this claim here.

  PublisherDOI

• Capturing Early Word Production in Caregiver-Toddler Shared Book-Reading Interactions

  Open MIND · 2026-01-01

  articleOpen accessSenior author

  In many circumstances, age and experience with language go hand in hand, making it challenging to disentangle their roles in early word learning. Here we use shared book-reading as a vehicle for structured but naturalistic language exposure, allowing us to control age of exposure to new words (i.e. novel nouns). We presented 18-month-olds (n = 24) and 22-month-olds (n = 20) with three novel nouns embedded in a picture book that caregivers read at home over two weeks. Caregiver-child reading sessions were audio-recorded and coded for toddlers’ novel noun production (our outcome measure), and properties of the reading sessions (such as reading session length and extratextual extensions beyond the book text). While 18-month-olds' and 22-month-olds' caregivers said the target words the same number of times during book-sharing, the older toddlers' caregivers engaged in longer reading sessions and used more extratextual extensions. Over the two weeks, nearly all 22-month-olds and over 2/3 of 18-month-olds produced at least one of the novel nouns (and most 22-month-olds produced all 3). Intriguingly, increased caregiver extratextual extensions were associated with a higher likelihood of toddlers producing novel words during reading sessions within each age group, and overall. In addition, toddlers’ overall productive vocabulary (based on a parent-report checklist, the MBCDI) predicted novel word production among 22-month-olds but not 18-month-olds. Together, these findings show that two weeks of naturalistic book-sharing (with about 250 exposures to each of 3 novel nouns) is sufficient to foster new word production. Our results reveal both developmental changes in toddlers’ learning capacities over 18 to 22 months, and the efficacy of certain aspects of caregiver input across this range. Caregivers’ extratextual talk during shared book-reading emerged as a particularly potent support for toddlers’ use of newly learned words even at the earliest stages of productive vocabulary development.

  PublisherOA PDFDOI

• Daylong patterns of object-centric interaction in two subsistence societies

  Infant Behavior and Development · 2026-05-12

  articleOpen access

  Object-centric interactions provide rich sources of multimodal input for early learning. Using daylong photo streams from child-worn cameras, we analyze > 90k images to identify the frequency and targets of object handling across the first three years of life in two small-scale subsistence farming communities on opposite sides of the globe (Tseltal Mayan and Rossel Papuan). We find that infant object handling is relatively frequent in both communities, capturing just under one-third of infants’ waking time, on average, and increasing with age. In contrast, others’ handling of child-relevant objects was exceptionally rare, capturing only 1–2% of infants’ waking time, on average, in both communities. The distribution of infants’ handled object categories (e.g., consumables, mealtime tools, natural kinds, etc.) was broadly similar across communities and relatively stable over development, likely reflecting consistent features of children’s physical environments and daily routines. However, the exact objects available to infants varied substantially both within and between communities. We discuss how the objects infants handle index their early experiences and offer a window into the sociocultural contexts shaping their everyday learning. • Daylong photo-linked audio recordings capture children’s everyday experiences in two rural non-Western communities. • Tseltal (Mayan) and Rossel (Papuan) infants under age three handled objects for just under ⅓ of their waking time. • Object handling rate increased with age across the first three years of life. • Distributions of handled objects were highly skewed within and across infants. • Interactant object handling was exceptionally rare in both communities.

  PublisherDOI

• Measuring early word exposure in infants: A brief parent-report survey captures individual language input and predicts vocabulary outcomes.

  Developmental Psychology · 2026-03-09

  articleOpen accessSenior author

  = 264, 78% White, 11% mixed-race, 11% other) with the same vocabulary and exposure surveys were consistent, supporting generalizability. These findings suggest a brief word exposure survey can reliably estimate individualized word input and knowledge, at least for highly common nouns. This method complements naturalistic recordings and corpus-based norms by capturing individual variation and enabling more scalable research across linguistic and sociocultural contexts. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

  PublisherDOI

• Response to Creel

  PsyArXiv (OSF Preprints) · 2026-04-15

  preprintOpen accessSenior author

  Our original paper (Liu, Hilton, et al., 2023) found effects of linguistic experience on music perception which generalized across a half-million speakers of fifty-four languages. In a critique, Creel (Creel, 2026) claims that our multilevel modelling approach biased the language-level estimates such that languages with small-sample sizes “spuriously appear to pattern consistently” compared to “raw” sample means. We respectfully disagree and counter this claim here.

  Publisher

• Examining spoken language input to infants with cochlear implants and its relationship to language outcomes

  2026-03-11

  articleOpen accessSenior author

  This study compared spoken language input to deaf/hard of hearing (DHH) children with cochlear implants, and two sets of hearing controls: age and hearing-age matches. Using daylong audio-recordings from 6-32 month-olds (n=16/group), we examined whether systemic variation in thirteen language input metrics predicted children’s emerging language outcomes. Language input to the DHH group contained shorter sentences and more auditory words (e.g., “loud”) than input to hearing-age-matches, and slightly fewer adult words than input to age-matches. Despite the modest input differences, DHH children produced fewer mature vocalizations than age-matched peers; their vocalizations also increased less robustly with age. Regression models indicated that age, hearing status, noise, input quantity, overlapping speech, and input MLU collectively predicted >50% of variance in children’s vocal maturity. These results suggest differences in children’s access to language input play a larger role than the relatively small systematic input differences across DHH and hearing children.

  PublisherDOI

• Capturing Early Word Production in Caregiver-Toddler Shared Book-Reading Interactions

  2026-04-21

  articleOpen accessSenior author

  In many circumstances, age and experience with language go hand in hand, making it challenging to disentangle their roles in early word learning. Here we use shared book-reading as a vehicle for structured but naturalistic language exposure, allowing us to control age of exposure to new words (i.e. novel nouns). We presented 18-month-olds (n = 24) and 22-month-olds (n = 20) with three novel nouns embedded in a picture book that caregivers read at home over two weeks. Caregiver-child reading sessions were audio-recorded and coded for toddlers’ novel noun production (our outcome measure), and properties of the reading sessions (such as reading session length and extratextual extensions beyond the book text). While 18-month-olds' and 22-month-olds' caregivers said the target words the same number of times during book-sharing, the older toddlers' caregivers engaged in longer reading sessions and used more extratextual extensions. Over the two weeks, nearly all 22-month-olds and over 2/3 of 18-month-olds produced at least one of the novel nouns (and most 22-month-olds produced all 3). Intriguingly, increased caregiver extratextual extensions were associated with a higher likelihood of toddlers producing novel words during reading sessions within each age group, and overall. In addition, toddlers’ overall productive vocabulary (based on a parent-report checklist, the MBCDI) predicted novel word production among 22-month-olds but not 18-month-olds. Together, these findings show that two weeks of naturalistic book-sharing (with about 250 exposures to each of 3 novel nouns) is sufficient to foster new word production. Our results reveal both developmental changes in toddlers’ learning capacities over 18 to 22 months, and the efficacy of certain aspects of caregiver input across this range. Caregivers’ extratextual talk during shared book-reading emerged as a particularly potent support for toddlers’ use of newly learned words even at the earliest stages of productive vocabulary development.

  PublisherDOI

• Measuring Early Word Exposure in Infants: A Brief Parent-Report Survey Captures Individual Language Input and Predicts Vocabulary Outcomes

  UNC Libraries · 2026-03-21

  articleOpen accessSenior author

  Individual differences in early language input are rampant but hard to quantify without resource-intensive naturalistic recordings, posing a challenge for testing links between individual experience and facets of language development like vocabulary growth. We tested whether a quick parent survey on infants' exposure to common nouns could reliably predict nouns' frequency in home recordings (testing its validity) and children's own vocabulary (testing input-learning links). In Study 1 (<em>n</em> = 44; 95% White, 5% mixed-race), we gathered monthly home recordings and exposure surveys every 2 months from 8 to 18 months. Parent-reported exposure to each noun (16/time point, 5-point scale) was compared to the noun's relative frequency in infants' home recordings and to comprehension and production of each noun via parent vocabulary survey (Communicative Development Inventory). Reported exposure significantly predicted noun frequency in home recordings. It also predicted both reported noun comprehension and production, even after accounting for age. Results from an age- and gender-matched cross-sectional sample (Study 2, <em>n</em> = 264, 78% White, 11% mixed-race, 11% other) with the same vocabulary and exposure surveys were consistent, supporting generalizability. These findings suggest a brief word exposure survey can reliably estimate individualized word input and knowledge, at least for highly common nouns. This method complements naturalistic recordings and corpus-based norms by capturing individual variation and enabling more scalable research across linguistic and sociocultural contexts.

  PublisherDOI

• Artificial intelligence–powered 3D analysis of video-based caregiver-child interactions

  Science Advances · 2025-02-14 · 9 citations

  articleOpen access

  We introduce HARMONI, a three-dimensional (3D) computer vision and audio processing method for analyzing caregiver-child behavior and interaction from observational videos. HARMONI operates at subsecond resolution, estimating 3D mesh representations and spatial interactions of humans, and adapts to challenging natural environments using an environment-targeted synthetic data generation module. Deployed on 500 hours from the SEEDLingS dataset, HARMONI generates detailed quantitative measurements of 3D human behavior previously unattainable through manual efforts or 2D methods. HARMONI identifies longitudinal trends in child-caregiver interaction, including child movement, body pose, dyadic touch, visibility, and conversational turns. The integrated visual and audio analysis further reveals multimodal trends, including associations between child conversational turns and movement. Open-sourced for large-scale analysis, HARMONI facilitates advancements in human development research. HARMONI achieves 63 to 80% consistency on key attributes with human annotators on SEEDLingS and 84 to 93% consistency on videos taken from a laboratory setting while achieving >100 times savings in time.

  PublisherDOI

• Comparing Utterance Types and Contents in the Input to Blind and Sighted Infants

  2025-08-24

  articleOpen accessSenior author

  Blind infants cannot acquire information about the world using vision, which may make it more difficult to connect words they hear to referents that, for sighted infants, are readily visually available. We ask whether, in light of this, caregivers talk differently to blind vs. sighted infants. We consider two qualitative dimensions of speech input: the types of utterances it contains, and the content of those utterances. Using daylong LENA audio-recordings, we captured naturalistic speech in the homes of blind and sighted infants (n=15 each, aged 6-30 months).We systematically subsampled and transcribed 40 minutes per child; these transcripts were then analyzed across various dimensions tapping e.g. syntactic structure, informativity, and conversational features. Overall, we found highly similar distributions of utterance types and content across the speech heard by the children in each group.Fine-grained analysis revealed blind children's language environments contained slightly but consistently more declaratives compared to sighted peers, with other categories showing no differences across the board. These findings show that parents in both groups talk to their children in similarly informative ways, challenging claims that blind infants hear more directives and commands.More broadly, this work expands on prior research suggesting that stark differences in caretaker input are unlikely to explain blind and sighted infants' early language development.

  PublisherOA PDFDOI

Recent grants

• CAREER: Understanding Early Language: Evidence from Blind Infants & Deaf Infants

  NSF · $466k · 2019–2023

• Mechanisms of Word Learning in Infancy

  NIH · $360k · 2014–2019

• Mechanisms of Word Learning in Infancy

  NIH · $1.6M · 2014–2020

Frequent coauthors

• Alejandrina Cristià

  Université Paris Sciences et Lettres

  79 shared

• Mélanie Söderström

  Stanford University

  37 shared

• Marisa Casillas

  University of Chicago

  34 shared

• Camila Scaff

  24 shared

• Marvin Lavechin

  22 shared

• Federica Bulgarelli

  University at Buffalo, State University of New York

  22 shared

• Anne S. Warlaumont

  University of California, Los Angeles

  18 shared

• Andrei Amatuni

  16 shared

Labs

• Elika Bergelson LabPI

Education

• Ph.D., Language & Mind

  University of Maryland

  2013

• B.A., Language & Mind, Music, and French

  New York University

Awards & honors

• Early Investigator Award
• NSF GRFP
• NSF IGERT