Statistical learning in children's emergent L2 literacy: Cross-cultural insights from rural Côte d'Ivoire

Abstract

Children around the world learn to read across radically different educational systems and communities. In the west African nation of Côte d’Ivoire, children enter the fifth grade (CM-1) with widely varying literacy skills in French, the official language for primary education. Previous studies have often linked performance in statistical learning tasks with differences in children’s and adults’ literacy outcomes, mainly in Western and high-income educational contexts. We asked whether Ivorian children’s individual differences in emergent second language literacy skills and analogous first language skills could also be explained by their performance in non-linguistic visual statistical learning (VSL). Across three iteratively-developed tasks, 157 children in rural communities surrounding the Adzópe region in Côte d’Ivoire completed a VSL task on touchscreen tablets. We found strong group-level evidence that children exploited the statistical regularities in the image sequence to decrease their response times, but in a post-test, their discrimination between valid and invalid sequences did not exceed chance. Overall, their responses were slower than U.S. children completing a similar task (Qi et al., 2019), and these individual differences in processing speed both confounded statistical learning and predicted second language emergent literacy skills. Moreover, the weak correlation to analogous skills in their first language further suggests that this task did not measure the same skills for the Ivorian children as in previous samples. We recommend adaptations to the statistical learning paradigm that may improve its generalizability across the wider global population.

Intact Procedural Memory and Impaired Auditory Statistical Learning in Adults with Dyslexia

“…This approach of measuring online learning has been validated in our previous work in adult learners: their responses to target stimuli accelerated more quickly in structured sequences, similar to the ones used here, than in random sequences where no triplets were formed and the same stimuli were displayed in a random order (Schneider et al, 2020;Tang et al, 2022). Significant RT acceleration was also observed in children using a tablet to respond to targets in the third position, but not the first position, of a triplet, indicating that they had learned the triplet structure and could anticipate the target (Zinszer et al, 2020).…”

Does non-linguistic segmentation still predict literacy in an L2 education? Statistical learning in Ivorian primary schools

“…We asked whether previously reported correlations between non-linguistic statistical learning (SL) and children's literacy existed for children in southeastern Côte d'Ivoire learning to read in their second language. Following extensive pilot research reported by Zinszer et al (2022) and in the supplemental materials, we proposed a set of adaptations to existing visual and auditory SL tasks to measure individual differences among children who (1) have no experience using personal computers, (2) are unaccustomed to repetitive forced-choice test designs, and (3) are located in rural communities with limited access to typical laboratory infrastructure. Our results provided moderate to strong evidence against direct correlations (outside the interval [-0.20, 0.20]) between SL and L2 literacy, but we identified relationships between French phonological awareness and visual SL and between French phonological awareness and L2 literacy.…”

“…Forty-seven children completed an auditory statistical learning task (ASL) using auditory pure tones, and 46 completed a visual statistical learning task (VSL) using cartoon aliens. The VSL data tasks were also reported by Zinszer et al (2022), but that paper used linear mixed-effects models while this analysis closely follows the data processing and analysis steps described by Qi et al (2019) for better comparison. Table 2 summarizes the experiment parameters.…”

“…In a pre-SL training phase, children performed a target detection task on a sequence of 72 audiovisual stimuli, composed of six cartoon monsters (see Zinszer et al, 2022 for images) paired with six environmental sounds. Trials were 800ms in duration (sound during the first 500 ms) with 200 ms interstimulus interval and randomly ordered according to a uniform distribution of transitional probabilities.…”

A meta-analysis on the correlations between statistical learning, language, and reading outcomes.

“…Note that some studies used adaptions of this test in languages other than English. Zinszer et al (2020) adopted the Woodcock–Johnson-III Test of Cognitive Abilities (Woodcock et al, 2001) to assess children’s vocabulary knowledge of native language (Attié) and their second language (French). Shafto et al (2012) asked parents to fill out the MacArthur–Bates Communicative Development Inventory (Fenson et al, 2006) to learn about infants’ vocabulary knowledge.…”