Stimulus familiarity shapes hierarchical structure learning and metacognitive dynamics

Humans learn multiple probabilistic regularities at once, and often without feedback, which may be aided by metacognitive evaluations throughout learning. To understand this learning-metacognition relationship, we must first characterize how both capacities co-evolve as environmental structures are internalized, especially when feedback is unavailable. Here, participants learned a two-level probabilistic environment in which stimulus co-occurrences were embedded within a sequence, using unfamiliar fractals and familiar line drawings in a two-session, within-subjects design (N=63; N=50 completed both sessions). For each stimulus set, we measured learning across six repeating exposure-test cycles, using a cover task during exposure phases and forced-choice questions in test phases; test questions also included confidence judgments. Results revealed stimulus- and structure-dependent learning and metacognitive signatures. First, learning dynamics differed across stimuli: fractal sequence test accuracy outpaced co-occurrence accuracy, but vice versa for line drawings. Further, confidence increased across blocks, and was higher for line drawings as well as co-occurrences and first but not second sequence selections. Finally, both metacognitive sensitivity and test accuracy increased in successive exposure-test cycles for fractals, but metacognitive sensitivity remained stagnant for line drawings even as test accuracy improved. These results demonstrate that structure learning and metacognitive processes are highly context- and complexity-dependent rather than being completely generalizable. This study informs future directions for refining theories of learning without feedback, including how people evaluate their own learning and uncertainty and the interaction of both cognitive abilities with task design, stimuli, and environmental structures.