A team led by Jina Ahn of the University of California, Los Angeles, together with Erica Cartmill of Indiana University and the Santa Fe Institute and Catherine Sandhofer of UCLA, ran two experiments with children between roughly two and five and a half years old. The design was elegantly simple. Each child first spent ten minutes in unstructured free play with a friendly adult experimenter, who followed the child’s lead, took turns with toys, and offered warm, encouraging comments. Then came the critical manipulation. Some children went on to learn new words from that same, now-familiar experimenter. Others watched her leave the room, only to be replaced by a complete stranger who delivered exactly the same lesson.

The lesson itself was a carefully engineered word-learning task. The researchers 3D-printed dozens of strange objects that no child had ever seen, painted in bright colors, and grouped them into categories of four exemplars sharing the same shape. Each category received a made-up label drawn from a validated database of novel names, words like Wiso, Toma, Bink, and Zeb that sound plausible in English but mean nothing. During training, the experimenter showed three exemplars of a category, labeled each one twice, and let the child handle them. Then, in a forced-choice test, the child had to pick the right object from four options: a new exemplar of the trained shape in a different color, a familiar but unlabeled distractor, and two objects never seen before. Choosing correctly required genuine generalization of the label to a new member of the category, not just recognition of a previously handled toy.

In the first study, thirty-six toddlers aged 22 to 37 months were randomly assigned to learn from either the familiar or the unfamiliar experimenter. The toddlers performed well above chance in both conditions, averaging just over four correct responses out of eight trials regardless of who taught them. The two group means were identical to two decimal places. To make sure this null result was not merely a failure to detect a real effect, the team turned to Bayesian statistics, a framework that quantifies how strongly the data support one hypothesis over another. Using a standard Cauchy prior, the analysis yielded a Bayes factor of about 3.1 in favor of the null hypothesis, meaning the observed pattern was roughly three times more likely if familiarity truly made no difference than if it did. That counts as moderate evidence that the ten-minute bond simply did not matter for word learning in this age group.

But there was a catch. The first study never directly measured whether the children actually preferred the familiar adult. Perhaps the warm-up session failed to create any bond at all, rendering the comparison moot. The second study was designed to close that loophole, and it produced the study’s most striking result. Thirty-four preschoolers aged 49 to 69 months met both experimenters: one familiar from the play session, one brand new. Immediately afterward, each child received a handful of star-shaped toys and two identical buckets, one belonging to each experimenter, with instructions to put all the stars into the bucket of the person they wanted to play with more. The choice was all-or-nothing, a direct expression of social preference.

The children voted overwhelmingly with their hearts. Twenty-six of the thirty-four, or 76.5 percent, gave every star to the familiar experimenter, a distribution far from the fifty-fifty split expected by chance. A mere ten minutes of positive interaction was enough to tip preschoolers’ social choices dramatically toward someone they had just met. This finding aligns with a rich literature showing that young children use resource allocation to signal affiliation, favoring familiar or socially favored individuals when fairness cannot be maintained, and that brief positive exchanges involving smiles, eye contact, and warm speech can powerfully shape whom children want to be around.

Yet when the same children sat down to learn words, the preference evaporated into irrelevance. Each preschooler completed four training-and-test blocks with the familiar experimenter and four with the unfamiliar one, learning different labels from each. They generalized above chance with both teachers, averaging 2.59 correct out of four with the familiar adult and 2.24 with the unfamiliar one. The difference was small and statistically indistinguishable from zero, and a paired Bayesian analysis, while numerically favoring no difference, fell just short of the conventional threshold for conclusive evidence. An exploratory model also found no link between which experimenter a child favored in the star task and how well that child learned from either teacher. Liking someone, in short, did not translate into learning more from them.

Why would familiarity reshape preference but not learning? The authors point to a fundamental asymmetry between imitation and word learning. Imitation demands close, continuous engagement with a partner: watching their hands, coordinating actions, reading their intentions in real time. In that context, comfort with a familiar person plausibly frees up attention and reduces wariness, and prior studies have indeed shown toddlers imitating familiar demonstrators more readily, but only when the demonstration was directed at them personally. Word learning, by contrast, can proceed with remarkable independence from the social relationship itself. Children can pick up new words simply by overhearing conversations between other people, and while speech directed at them is more effective, the mapping of sound to meaning appears robust to who happens to be speaking, provided the linguistic input is clear.

The researchers are careful about the limits of their claim. Both studies delivered explicit, unambiguous labels in crystal-clear referential contexts, which may have left little room for social factors to matter. Familiarity could yet prove decisive in messier situations: when reference is ambiguous, when two speakers give conflicting information, or when a child must infer what a speaker intends from subtle pragmatic cues. The testing also occurred immediately after training and measured only receptive generalization, so effects on retention, consolidation, or production remain unexplored. And because children were tested in their own familiar preschools, even the unfamiliar experimenter may not have felt truly strange, potentially muting the manipulation. Temperament matters too; shy or behaviorally inhibited children might respond very differently to a stranger.

Still, the takeaway is quietly revolutionary for anyone who designs educational settings for young children. The warm-up period before a lesson, the friendly face at story time, the rapport-building that teachers instinctively do: these may matter enormously for how children feel, whom they choose, and how willing they are to engage, without conferring any measurable boost on structured vocabulary learning itself. Children, it seems, can separate the social from the informational. They will happily give their stars to the person they played with, and then learn just as readily from the stranger standing beside them. In a world where parents fret about finding the perfect tutor and policymakers debate teacher quality, the study suggests that for the youngest learners, the quality of the linguistic input may matter more than the warmth of the messenger, at least once that input is clear enough to speak for itself.

Subject of Research: Developmental Psychology

Article Title: Short-term familiarity shapes social preference but not word learning in preschoolers

Article References: Ahn, J., Cartmill, E. A., & Sandhofer, C. M. (2026). Short-term familiarity shapes social preference but not word learning in preschoolers. Frontiers in Developmental Psychology, 4, Article 1895134. https://doi.org/10.3389/fdpys.2026.1895134

Image Credits: AI Generated

DOI: 10.3389/fdpys.2026.1895134

Keywords: child development, word learning, social preference, familiarity, language acquisition, Bayesian statistics, Short-term, shapes, social, preference, word, learning