Children’s distributional learning of recursive structures: Capacities and constraints

This study investigates how children learn recursive structures, a core property of human language. While recursion is considered universally available, languages differ in the specific rules governing recursive embedding, which poses a learnability challenge. Previous research has shown that adults can learn the recursivity of a structure in an artificial language from distributional cues in non-recursively embedded data, and this learning mechanism is constrained by the high-level structural representation. However, it remains an open question whether children can do the same. We addressed this question in two artificial language learning experiments. In Experiment 1, we found that children, like adults, can use distributional cues in non-recursive input to acquire recursive structures. In Experiment 2, we found that while children also learned the headedness of the structure from distributional cues, only older children were able to integrate this information to constrain the licensing of recursion. These findings suggest that children can acquire complex grammatical knowledge through distributional learning, and that their distributional learning ability changes over the course of development, particularly the ability to integrate multiple sources of information.