Abstract
Relational rules organize simple associations into structured knowledge, enabling generalization to novel contexts — a cornerstone of reasoning in domains like mathematics. To investigate how such rules are learned, we used an analog of mathematical inequalities. The full inequality rule was decomposed into three cases. Across a series of 2×2 experiments, we manipulated training format (interleaved vs. blocked) and example exposure (present vs. absent). As predicted by previous studies, blocked training yielded better performance on the full task than interleaved training, such that blocking lends itself to the discovery of complex relational structures. These results suggest that the structure of training, rather than exposure to examples, plays a critical role in relational rule learning.