We agree with Lew-Levy and Amir that peer culture is a vital yet underexplored aspect of children’s early learning environments. However, we argue that the importance of peer language input has been systematically overlooked, particularly before age three, due to limited means of capturing naturalistic interactions between children. But with the emergence of tools for processing long-form audio recordings, which offer a discrete and hands-off way of studying peer cultures, this barrier is being overcome.

Long-form recordings refer to a suite of methods for analyzing naturalistic audio recordings of everyday events collected via child-worn recording devices. Tools like Language Environment Analysis (LENA) (Gilkerson et al., Reference Gilkerson, Richards, Warren, Montgomery, Greenwood, Kimbrough Oller and Paul2017) and Voice Type Classifier (VTC) (Lavechin et al., Reference Lavechin, Bousbib, Bredin, Dupoux and Cristia2020) allow researchers to distinguish between child, adult, and peer speech from other sources of noise with increasing accuracy, even in multilingual and under-resourced settings (e.g., Cristia et al., Reference Cristia, Gautheron and Colleran2023; Coffey et al., under review). These systems leverage new advances in automated speech analysis, allowing researchers to collect and process massive amounts of speech data from nearly any setting. These features greatly expand the possible scope of language studies and offer new answers to major assumptions in early linguistic development. For instance, a study of over 38,000 hours of home recordings in the U.S. suggested that the word gap between high- and low-income families is far smaller than previously estimated (4 million words by 4 years per Gilkerson et al., Reference Gilkerson, Richards, Warren, Montgomery, Greenwood, Kimbrough Oller and Paul2017 vs. 30 million per Hart & Risley, Reference Hart and Risley1995). Another large-scale cross-cultural study over 40,000 hours of audio taken from families in 12 countries found little evidence of socioeconomic or gender differences in the rates that young children vocalize (Bergelson et al., Reference Bergelson, Soderstrom, Schwarz, Rowland, Ramírez-Esparza, Hamrick and Cristia2023).

These methods are potentially transformative for the study of peer culture. As the authors note, peer cultures are not only distinct from those of adults but also tend to “insulate” themselves from adult presence. Traditional research methods, such as in-person observation, interviews, lab-based experiments, and non-portable recording devices, have failed to capture their naturalistic everyday dynamics. Much of what we know about children’s language environments has been captured in the presence of adults. Studies suggest that children are more performative around adults and vocalize more in lab settings than in natural environments (Lewedag, Oller, & Lynch, Reference Lewedag, Oller and Lynch1994; Bergelson et al., Reference Bergelson, Amatuni, Dailey, Koorathota and Tor2019), and much of their time is spent away from adult eyes (Davies & Cashdan, Reference Davis and Cashdan2020). Our personal interactions with research participants support these claims. For instance, a Tsimane mother from the Amazon laughed upon hearing a recording of her 4-year-old child pretending to be a pilot during a game with his siblings. She explained she had never heard him speak that way – he never told those kinds of stories to her. As this moment exemplifies, much of what we know about child development has been filtered through the “adult gaze,” effectively obscuring the very peer-to-peer input that appears so critical in shaping children’s social and linguistic growth.

With these systems, researchers have been able to shed new light on the uniquely peer-driven aspects of language development that have gone unnoticed in traditional observational studies. Studies using long-form recordings in small-scale subsistence-based societies have routinely found that young children hear more speech input from other children than adults as they age and that children’s vocalizations are better predicted by input from other children than from adults, contrasting sharply with findings from societies where sibling caregiving is less common (Weber Fernald, & Diop, Reference Weber, Fernald and Diop2017; Casillas, Brown, & Levinson, Reference Casillas, Brown and Levinson2021; Cristia, Gautheron, & Colleran, Reference Cristia, Gautheron and Colleran2023; Scaff et al., Reference Scaff, Casillas, Stieglitz and Cristia2024; Coffey et al., under review). While transcribed studies of peer input in these settings were limited in the number of children that could be observed (e.g., n = 15 in Shneidman & Goldin-Meadow, Reference Shneidman and Goldin-Meadow2012; n = 3 in Loukatou et al., Reference Loukatou, Scaff, Demuth, Cristia and Havron2022), these systems have massively increased the data available to study, from n = 24 in Scaff et al. (Reference Scaff, Casillas, Stieglitz and Cristia2024) to n = 276 in Coffey et al. (under review). In urban Western settings, long-form recordings have been used in classroom settings, where many of these trends have also been observed: children talk more amongst each other than they do their teachers, and their speech production is better predicted by peer input than adult input (Dykstra et al., Reference Dykstra, Sabatos-DeVito, Irvin, Boyd, Hume and Odom2013; Perry et al., Reference Perry, Prince, Valtierra, Rivero-Fernandez, Ullery, Katz and Messinger2018).

These methods are widely available to researchers looking to supplement their research, not just limited to language studies but also as a proxy for interaction or social networks. LENA is available through the LENA Foundation with specialty programs for home-based or classroom-based research. VTC is available through ChildProject, an open-source collection of tools for analyzing long-form recordings, consisting of over 40 data sets, drawing from over 1000 children across 6 continents and dozens of languages (Gautheron, Rochat, & Cristia, Reference Gautheron, Rochat and Cristia2023). Although long-form recordings cannot tell researchers the content about what is said between peers, software analysis allows for selective sampling of speech from different sources or from periods of high volubility for further manual annotation (Casillas, Brown, & Levinson, 2020; Reference Casillas, Brown and Levinson2021; Bunce et al., Reference Bunce, Soderstrom, Bergelson, Rosemberg, Stein, Alam and Casillas2025). There are also tools in development that use these systems to identify segments of speech to deploy automated transcription of child speech (e.g., Radford et al., Reference Radford, Kim, Xu, Brockman, McLeavey and Sutskever2023; Chaparro-Moreno et al., Reference Chaparro-Moreno, Villasanti, Justice, Sun and Schmitt2024; Pelfrey et al., Reference Pelfrey, Justice, Gonzalez Villasanti and Foster2024; Sun et al., Reference Sun, Londono, Elbaum, Estrada, Lazo, Vitale and Messinger2024). Thus, long-form recordings can be used as a viable alternative to direct observation as a method of collecting naturalistic speech, which is key to understanding peer speech practices, as well as the innovation of novel words and linguistic forms.

Long-form recordings are a powerful solution to a blind spot in developmental science: the systematic neglect of peer language input, particularly before age three. By capturing naturalistic, peer-driven speech across diverse settings, long-form recordings reveal that much of children’s language development occurs beyond adult observation. By vastly increasing the kind and amount of data available to researchers, these systems will play a vital role in unlocking peer culture.