ManyBabies 3: A Multi-Lab Study of Infant Algebraic Rule Learning

Visser, Ingmar and Geambașu, Andreea and Kachergis, George and Oliveira, Catia M and Rocha-Hidalgo, Joscelin and Zettersten, Martin and Ahamat, Teruni Malaika and Alessandroni, Nicolás and Althaus, Nadja and Arunachalam, Sudha and Aussems, Suzanne and Axelsson, Emma and Aydin, Zeynep and Baumgartner, Heidi A and Bergmann, Christina and Bettoni, Roberta and Bulf, Hermann and Byers-Heinlein, Krista and Capparini, Chiara and Carroll, Lily and Carstensen, Christian Alexander and Cebulla, Rebekka and Chen, Xiaoyun and Chladkova, Katerina and Colavolpe, Benedetta and Constantine, Rodica and Cortes, Elisabet Eir and Doyle, Frances and Exner, Anna and Black, Alexis and Forbes, Samuel H. and Franchin, Laura and Geraci, Alessandra and Gervain, Judit and Hagelund, Silje and Hannon, Erin and Havron, Naomi and Jaffe-Dax, Sagi and Johnson, Scott and Jędryczka, Wiktoria and Kartushina, Natalia and Struhl, Melissa Kline and Koleini, Ada and Kona, Abhini and Kong, Shannon and Kosie, Jessica Elizabeth and Kučerová, Monika and Kynclova, Katerina and Leslie, Jared and Lew-Williams, Casey and Li, Scarlet Wan Yee (2025) ManyBabies 3: A Multi-Lab Study of Infant Algebraic Rule Learning. Other. PsyArXiv.

Abstract

The ability to learn and apply rules lies at the heart of cognition. In a seminal study, Marcus et al.(1999) reported that 7-month-old infants learned abstract rules over syllable sequences and wereable to generalize those rules to novel syllable sequences. Dozens of studies have since replicatedthat finding and extended it using different rules, modalities, stimuli, participants (human adultsand non-human animals) and experimental procedures. Yet questions remain about thegeneralizability of Marcus et al.’s (1999) core findings, and sources of variation across thesefindings. In the current study, we address this issue by testing 839 infants of a wide age range(5;0-12;0 months) in a multi-laboratory (30 laboratories, [31 samples]) conceptual replication ofthe Marcus et al. (1999) study. This study and the analyses were submitted as a registered report.A mixed-model analysis of the looking times at consistent vs. inconsistent test trials indicated nosignificant effect. The current study therefore finds no evidence for rule-learning; to the contrary, aBayesian analysis of the data indicated strong evidence for a null effect. This was similarly thecase for the moderating factors that were part of the design: mono- vs. multilingualism, age, andexperimental paradigm; none of these factors showed significant effects. Robustness analysesindicated that the results were not different in the case of different choices in the (pre-)processingof the data. A follow-up analysis for single labs indicated that there was no evidence ofrule-learning in 30 of the 31 contributed samples. The current finding raises fundamental issuesabout rule-learning that are presented in Discussion.