About this paper symposium

---

Panel information
Panel 11. Language, Communication

Paper #1
Does curiosity predict learning in infants?
Author information	Role
Dr. Marina Bazhydai, Ph.D., Lancaster University, United Kingsom	Presenting author
Elena Constanze Altmann, Georg-August-Universität Göttingen, Germany	Non-presenting author
Malcolm K. Y. Wong, Lancaster University, United Kingdom	Non-presenting author
Gert Westermann, Lancaster University, United Kingdom	Non-presenting author
Abstract
Research on active learning has demonstrated that having control over informational input enhances attention and memory and leads to faster and more robust learning in adults and older children (e.g., Gureckis & Markant, 2012). However, emerging evidence from early childhood has produced inconclusive results (Ackermann et al., 2020; Fandakova & Gruber, 2020; Pereira et al., 2014; Partridge et al., 2015; Ruggeri et al., 2019; Zettersten & Saffran, 2020). Specifically, it remains unknown whether enabling infants to actively control their learning process leads to better retention of novel labels. The current study investigated 20-23-month-old infants’ (N = 75) curiosity-driven information selection in a novel word learning task, designed to identify any potential advantage for active learning over passive learning. The gaze-contingent eye-tracking experimental paradigm capitalized on infants’ gaze rather than manual actions as an expression of curiosity, using saccadic eye movements as visual information sampling. Infants in one (Curiosity) condition were given the opportunity to structure their own information seeking to actively create word learning opportunities for themselves, while infants in two other conditions (Yoked and Random) engaged in learning novel words passively. Infants’ learning of word-object associations (baseline-corrected proportion of cumulative looking time to the target object at test) was compared across active and passive learning paradigms. Contrary to the hypothesis, the results indicate no advantage of active information selection on retention of novel words above and beyond passive learning; F(2, 72) = 0.11, p = .900, BF01 = 8.065 (See Figures 1 and 2). Interestingly, infants across all conditions retailed novel words above chance (ps <.05). This study advances our understanding of early word learning, demonstrating the resilience of infant word learning mechanisms, and warrants further investigations of the role of curiosity in different learning contexts and with more fine-grained research questions. For instance, the self-generated sequences themselves could be investigated (Altmann et al., 2024; Garon & Doucet, 2024), with recent work showing that clustering in parents’ toddler-directed speech results in superior word learning during natural play (Slone et al., 2022). Using a similar gaze-contingent paradigm with 13-15-month-olds, ongoing research collaboration (Altmann, Bazhydai, Westermann, & Slone, in progress) is testing information encoding benefits of experimentally varied timing schedules (e.g., massed vs. spaced vs. clustered information presentation) and compare those to infants’ self-generated sampling sequences. Overall, these studies will lead towards a more nuanced understanding of the mechanisms of active, infant-directed, curiosity-driven learning in infant learning.

Paper #2
Preschoolers and adults seek out talker variability during word learning
Author information	Role
Federica Bulgarelli, Ph.D., University At Buffalo, United States	Presenting author
Marina Rabideau, University at Buffalo, United States	Non-presenting author
Abstract
Words sound different every time they are said, both by the same person and across people (Bulgarelli et al., 2021). This variability has been found to have mixed effects on learning. For example, 14-month-olds benefit from talker variability when learning minimal pairs, but struggle when learning distinct-sounding words (Höhle et al., 2020; Bulgarelli & Bergelson, 2023). Additionally, while adults are slower to process speech when the talker changes (Mullennix et al., 1991), preschoolers and adults benefit from talker variability for learning to produce novel words (Richtsmeier et al., 2009; Barcroft & Sommers, 2005). These previous studies have used passive exposure, presenting participants with predetermined orders or fixed types of variability. Here, we ask whether adults and preschoolers seek out or avoid talker variability for learning new words. Using an active learning paradigm (e.g. Zettersten & Saffran, 2021; Foushee et al., 2023), participants choose who to learn words from. Then, we test 1) whether participants seek out variability during word learning and 2) how these choices influence word learning. Adult participants are 56 (target sample = 60) students recruited from the Psychology Subject Pool. Preschool participants are 18 (target sample = 40) children (mean age = 4). Participants played a game during which they learned new word-object pairs. Adults learned 11 Lithuanian names for familiar objects (Sinkeviciute et al., 2019); preschoolers learned 6 novel names for novel objects (from the NOUN database). In the Active condition, participants chose which of 8 speaker icons to hear the word from, each word was heard 8 times. Participants were then tested on production and comprehension, see Figure 1. Adults selected 5.33 speakers per word (range 1.54-8). Adults exhibited significantly above chance comprehension accuracy (M = 52%, sd = 24; p <.001). Participants’ production accuracy was 22% (sd = 41), which was also significantly above chance (p<.001). Comprehension and production accuracy were highly correlated with each other (r = .72, p<.001). The average number of speakers chosen was not significantly related to comprehension or production accuracy (p <.05). Preschoolers selected 5.97 speakers per word (range 1.66-8). Preschoolers exhibited significantly above chance comprehension (M = 38%, sd = 15, p<.001). They rarely produced responses during the production test, only 5/86 (6%) productions were correct. For preschoolers, choosing more speakers was negatively correlated with comprehension accuracy, though not significantly with this subset of the planned sample (r = -.41, p = .089), see Figure 2. Our results suggest that learners do not avoid talker variability during learning, but that choosing more or less variability may not necessarily benefit learning. However, adults’ learning here was lower than that previously found using the same stimuli, suggesting that active information seeking may have changed task demands. Next, we will collect data with both age groups in a Passive condition, wherein participants will receive the choices made by a yoked participant from the Active condition. This will allow us to tease apart effects of variability from active information seeking. For example, more variability may have a stronger effect on learning in a passive condition.

Paper #3
Goldilocks and the Three Books
Author information	Role
Ruthe Foushee, Ph.D., New School for Social Research, United States	Presenting author
Azzurra Ruggeri, Central European University Vienna, Technical University Munich (TUM) School of Social Sciences and Technology, Austria	Non-presenting author
Abstract
A core concept in developmental psychology and education is that children learn most effectively when exposed to stimuli that are moderately challenging: neither too simple—offering nothing new to learn, nor too complex—exceeding their current understanding. Though highly intuitive, this idea has been remarkably difficult to test. Previous research has primarily been conducted (1) in laboratory settings, (2) with infants (though see Cubit et al., 2021), and (3) using highly simplified visual stimuli (though see Gerken et al., 2011; Kidd et al., 2014). In this project, we aimed to examine children’s preferences for different levels of real-world language complexity, and how those preferences impact learning across a broader age range and in a more ecologically valid context: the classroom. Method (See Figure 1). Participants were 254 children (3.12–9.75 years, M=6.22), tested in a quiet area at their school or preschool. An experimenter read them the first three pages of three children’s books, written at three different levels of complexity (manipulated via words’ estimated ages of acquisition, or AOA). Children selected the book they wanted to continue hearing. The book that children actually heard to the end was determined by random assignment to one of three conditions: Simple (n=84), Intermediate (n=85), or Complex (n=85). After the story, children were tested on their learning of five novel words embedded in the narration. We focus our analysis on the subset of our sample (n=193) who also completed the Peabody Picture Vocabulary Test (PPVT; Dunn & Dunn, 2007). Our data allow us to examine whether children are sensitive to language complexity: (a) when making an explicit choice, and (b) in how effectively they learn. Results: Do children choose the ‘just right’ book? Children chose the three books at similar rates (Simple: 38%; Intermediate: 31%; Complex: 31%) overall. However, a multinomial logit model on children’s choices suggests that, as children’s linguistic sophistication (PPVT) increased, they became significantly more likely to select the Complex book, relative to the Simple book (Odds Ratio=1.47 [1.03, 2.10], p=0.033; Figure 2). Do children learn better from the ‘just right’ book? Children learned best from the Simple book (M=0.64 [0.60, 0.68]), relative to the Intermediate (M=0.55 [0.51, 0.60]) or Complex (M=0.53 [0.48, 0.58]) books. A logit model predicting children’s test accuracy from condition and child PPVT (standardized) confirms a significant effect of condition (Intermediate OR: 0.62 [0.45, 0.86]; Complex OR: 0.57 [0.41, 0.79]; X2(2)=13.2, p<0.001), and of PPVT (OR=1.38 [1.21, 1.58]; X2(1)=23.3, p<0.001), such that children with greater pre-existing language knowledge showed greater learning of the novel words embedded in the stories’ narrations. Our results speak to the compounding effects of earlier learning in facilitating later learning. More generally, our study models bringing classic developmental psychology premises out of the lab and into the real world.

Paper #4
How children seek information to learn the extension of novel words
Author information	Role
Jamie Chen, Princeton University, United States	Presenting author
Casey Lew-Williams, Princeton University, United States	Non-presenting author
Martin Zettersten, University of California, San Diego, United States	Non-presenting author
Abstract
When learners first encounter a word (like “dog”) together with a novel referent (e.g., a Dalmatian), they face ambiguity about what category the new word may generalize to (Xu & Tenenbaum, 2007). For example, “dog” might refer to Dalmatians (subordinate category level), dogs in general (basic level), all animals (superordinate level), or an even more general category like ‘all living things’ (hypernym level). How do learners generalize word meanings? One solution is that learners actively seek new information to clarify the extent of meaning. For example, they could sample new referents at different category levels, and test whether the new word also applies to Labradors vs. broader categories. In a preregistered experiment (https://osf.io/qsfdy), we investigated how 5-8-year-old children (N=83; M=6.9 years, range: 5.1-9.0 years) actively seek information when learning about a novel word’s extension. A word learning task closely modeled past studies of word learning at different levels of categorical abstraction (Lewis & Frank, 2018; Figure 1), in which participants learned a new word (e.g., “sibu”) that was associated with a single exemplar (1-item training condition; e.g., Dalmatian), or a set of three exemplars (3-item training condition) that belonged to the same subordinate (e.g., three Dalmatians), basic (three dogs), or superordinate category (three animals). Learners could then choose a novel exemplar in the Sampling Phase. On choosing an item, children learned whether the selected exemplar shared the same word (i.e., was or was not a sibu). Children were always presented with four sampling options: subordinate-, basic-, superordinate-, or hypernym-level items (e.g., vegetable). In the subsequent Test Phase, children were asked to generalize the word to a new set of exemplars at varying category levels. First, we replicated the “suspicious coincidence” effect: children were less likely to generalize a novel word to the basic level when being trained on three subordinate exemplars compared to just a single exemplar (logistic mixed-effects model; z=-3.39, p<.001). Moreover, training condition substantially modulated children’s choices at test in general. Training condition was a strong predictor of basic-level (chi-squared(3)=66.21, p<.001) and superordinate-level choices (chi-squared(3)=124.06, p<.001), showing that children were sensitive to the training sample when making generalizations (Figure 2). Contrary to our preregistered prediction, children did not shift their sampling choices systematically in response to different training conditions (chi-squared(6)=1.47, p=.96). However, children’s sampling became more exploratory with age: older children were more likely to sample an exemplar that went beyond the current training condition (e.g., selecting a Labrador after being trained on three Dalmatians; z=2.05, p=.04; see Figure 2). Counter to some proposals that children are more exploratory at younger ages, our results suggest that when children generalize word meanings, they are initially more likely to confirm their hypotheses than to explore. These findings have important implications for understanding how children’s information-seeking strategies vary across learning contexts and what mechanisms support generalizing a word’s meaning.

⇦ Back to session