Statistical learning performance is impacted by a previous learning experience: A predictive eye-movement study

Statistical learning (SL), the ability to extract recurrent patterns from sensory input, plays an important role in a range of cognitive functions. While much research has studied SL in stable artificial environments, natural inputs are rarely fixed: regularities are often probabilistic and continuously changing. A key question, therefore, is how SL unfolds under such conditions and to what extent it is shaped by learners' previous experiences. In the present study, we asked how trajectories of predictability ranging from highly structured to noisy sequences impact SL performance, and how learning in such conditions affects subsequent learning. To do so, we created a "Whack-a-Mole" game in which mole locations partially predicted subsequent mole locations, while the extent of predictability differed between blocks. In Experiment 1, predictability of mole locations increased across blocks in the first session and decreased in the second session, or vice versa. In Experiment 2, predictability followed the same trajectory in both sessions (either decreasing or increasing). Learning performance was measured using both reaction times and predictive eye movements toward target locations. Across studies, our findings reveal that learning in the second session was shaped by prior experience in the first session. Starting the second session with high predictability facilitated learning, whereas starting with low predictability hindered it, even when predictability later increased. These results suggest that learners are not passive absorbers of regularities but active information seekers, whose expectations about environmental structure impact learning. We discuss the implications of these findings for theories of SL in dynamic environments.