Children's narrow learning bottleneck accelerates the emergence of statistical properties of language

Learners are limited in how much they can learn when a language is passed from one generation to the next. This transmission bottleneck is thought to play a crucial role in shaping language by promoting the emergence of systematic structure that enhances learnability. Prior work has investigated the impact of transmission bottlenecks by experimentally manipulating how much input learners receive. However, in natural language learning, bottlenecks can also result from learners' cognitive limitations. Here, we take a naturalistic approach to testing the effects of such learning bottlenecks by comparing cultural transmission in two populations that differ in their cognitive capacities: children and adults. Specifically, we investigate the emergence of two cross-linguistic statistical properties that were shown to facilitate learning: statistically coherent units, and Zipfian frequency distributions. Transmission chains of child and adult learners observed and reproduced sets of colour sequences that were produced by the preceding participant, simulating the process of cultural transmission. For both age groups, initially random sets of sequences evolved to have statistically coherent units with a Zipfian frequency distribution, but crucially, these statistical properties emerged faster with children. In addition, both children and adults learned the final child-generated sequence sets more easily than the adult-generated ones. Together, these findings suggest that children's more limited cognitive capacities create a narrower learning bottleneck, which in turn accelerates the emergence of statistical properties that enhance learnability. These results provide novel evidence that cognitive constraints can modulate the emergence of systematic structure that facilitates learning.