Pilot study maps preschool word retention across speaker reliability and language background

A small online pilot study of 39 children aged 4 to 6 found that the apparent influence of a speaker’s reliability on children’s retention of novel labels changed over time. Among 21 monolingual children, performance was above chance immediately after teaching for both accurate and inaccurate speakers. After five minutes, only labels taught by the accurate speaker remained above chance. By 24 hours, the pattern had reversed: monolingual children performed above chance for the inaccurate speaker but not for the accurate speaker. Results from 18 English-Spanish bilingual children were less orderly, with no significant effects of speaker condition, testing time, or their interaction in the mixed-effects analysis. Because the study included small, unequal groups and was not designed for direct between-group comparisons, the findings characterize patterns within each population rather than establish a reliable difference between them.

The experiment used a selective-trust design that tested memory and social evaluation together. During a familiarization phase, children met two unfamiliar adult women shown on screen. One consistently named four familiar objects correctly, while the other consistently gave incorrect names, such as calling a ball a pen. The researchers then taught each child four nonsense labels for two unfamiliar objects. Both speakers supplied a label for each object, and each object-label pairing was presented twice during teaching. A third, unrelated speaker later asked children to identify the object named by a target label. Children demonstrated retention by selecting the correct object in a two-alternative forced-choice task. The paradigm therefore measured whether children remembered a word-object association, rather than whether they endorsed a speaker’s general reliability or preferred that speaker’s label.

Testing occurred at three points: immediately after instruction, five minutes later and 24 hours later, with a permitted one-hour window around the final session. During the five-minute interval, an experimenter played an online “I Spy” game with the child. Each of the four labels was tested three times at each point, producing 12 trials per time point and 36 trials in total. The same two unfamiliar objects appeared throughout, although their left-right screen positions and trial order were varied. This repeat-testing structure made it possible to compare retention over time with the same children, but it also means that elapsed time cannot be separated cleanly from repeated exposure to the objects and prompts. The investigators therefore measured whether accuracy changed across testing occasions, rather than isolating a pure effect of forgetting over time.

The final sample comprised 21 children classified as monolingual and 18 as bilingual. Monolingual participants had been exposed only to English from birth, with no more than 5% weekly exposure to another language. The bilingual group consisted of simultaneous English-Spanish learners exposed to both languages by 36 months and without significant exposure to a third language. Participants were recruited through convenience sampling in the United States, using online advertisements, flyers and word of mouth. Fifteen of 54 recruited children were excluded: four had difficulty attending to and following instructions in their home environment, while 11 could not provide sufficient usable data because of technical or completion problems involving device compatibility, audio, internet connectivity or screen sharing. The remaining children completed all three test points across sessions on consecutive days, with each session lasting less than an hour.

Within-group analyses produced the clearest results for monolingual children. At the immediate test, median accuracy was 0.67 for both the accurate and inaccurate conditions, and each was significantly above the 0.50 chance level. Five minutes later, accurate-speaker labels had a median accuracy of 0.75 and remained above chance, whereas inaccurate-speaker labels had a median of 0.50 and did not differ reliably from chance. At 24 hours, the accurate condition fell to a median of 0.50 and was not above chance, while the inaccurate condition reached a median of 0.75 and was above chance. A logistic mixed-effects model confirmed a condition-by-time interaction for the comparison between the immediate and 24-hour tests. At 24 hours, the odds of a correct response were lower in the accurate condition than in the inaccurate condition, with an odds ratio of 0.32 and a 95% confidence interval of 0.14 to 0.75. The immediate and five-minute comparisons did not show significant differences between the two speaker conditions.

The bilingual results did not reveal a similarly systematic trajectory. Children were above chance for labels from the inaccurate speaker immediately after teaching and after five minutes, with median accuracy of 0.67 and 0.50, respectively. Their accurate-speaker performance was not above chance at either point or at 24 hours, and neither condition was above chance at the final test. However, the mixed-effects model found no significant main effect of speaker condition, no significant main effect of testing time and no significant interaction between them. The appropriate conclusion is therefore not that bilingual children behave in a particular alternative way, but that this modest and heterogeneous sample did not yield a detectable structured pattern. Unequal group sizes also mean that a significant result in the monolingual group cannot be interpreted as a meaningful group difference.

The unexpected reversal after 24 hours requires caution. It was not one of the study’s predictions, which anticipated that accurate-speaker labels would be more likely to persist than inaccurate-speaker labels at delayed tests. The researchers explicitly state that sampling variability cannot be ruled out in a pilot sample of this size. One candidate explanation is that inaccurate labels, being more surprising, may receive more distinctive encoding and later become easier to retrieve. Another possibility is that the online format weakened social-evaluative information, allowing children to treat the inaccurate speaker’s words as arbitrary object names rather than as statements from an untrustworthy source. Delayed testing also places greater demands on retrieval, and repeated presentation of the same test items may have contributed to the result. These ideas remain hypotheses for adequately powered studies, not mechanisms demonstrated by the experiment.

The study also demonstrates both the promise and the practical difficulty of multi-session research with young children at home. Online testing allowed families throughout the United States to take part and yielded analyzable data at every scheduled point for the children who completed the protocol. At the same time, 27.8% of recruited children were excluded, including 11 because of technical or completion problems and four because home distractions impaired participation. The authors estimate that a study targeting a specific analyzable sample might need to recruit roughly 40% more children. Technology checks for device compatibility and audio, along with guidance to caregivers about reducing distractions, could reduce losses in future remote studies. The bilingual sample also varied in language exposure, suggesting that future work should measure such backgrounds more precisely rather than treating bilingualism as a single uniform condition.

Several design constraints limit how far the results can be generalized. Only two unfamiliar objects and four labels were used, and item-level random effects could not be estimated because the statistical models failed to converge when those effects were included. The inaccurate speaker was consistently wrong, so the experiment cannot show whether partial inconsistency would produce a different pattern. The repeated test items, variable home settings and technical failures add further uncertainty, while convenience recruitment restricted participation to families with suitable internet access, devices and time for two sessions. The study is best understood as an initial extension of an established speaker-accuracy paradigm into multi-session retention testing, not as a direct comparison of bilingual and monolingual learning. Larger, more diverse samples with direct between-group testing, finer measures of language experience, varied speaker consistency and experimental controls for repeated exposure would be needed to determine whether the apparent reversal after a day is reproducible and what mechanism produces it.

Subject of Research: Effects of speaker reliability, retention interval and language background on preschool children’s novel-word retention

Article Title: Selective word learning in monolingual and bilingual children: a retention study

Article References: Gangopadhyay, I., Lane, G., & Banerjee, S. (2026). Selective word learning in monolingual and bilingual children: a retention study. Frontiers in Language Sciences, 5, 1917744. https://doi.org/10.3389/flang.2026.1917744

Image Credits: AI Generated

DOI: 10.3389/flang.2026.1917744

Keywords: word learning, speaker accuracy, selective trust, bilingual children, monolingual children, memory retention, preschool development, online research, English-Spanish bilinguals, Selective, word, learning