During conscious learning, observers prioritize which aspects of the input to learn in an efficient, goal-directed manner. Statistical learning (SL) unconsciously extracts and represents patterns of individual elements, or “chunks,” from the environment. However, it remains unknown whether this automatic processing also follows a biased learning strategy, as higher-level learning does.
We hypothesized the existence of general chunking principles in SL that operate at the lowest perceptual level across sensory attributes and modalities when extracting statistical regularities. To test this, we relied on the Iambic–Trochaic Law (ITL), which posits that in auditory sequences, an element with longer duration signals the end of a segment to the observer, leading to decreased accuracy in detecting perceptual changes at segment boundaries.
We implemented the same stream-segregation paradigm in audition and vision to explore the generality of the ITL. Simple tones or basic visual objects of 200 ms and 600 ms duration were presented in a stream of three-element repeating pattern in a go/no-go paradigm with the task of detecting changes in gap duration. Participants exhibited similar boundary-related sensitivity changes in both modalities, confirming the “longer-last” principle. These findings suggest modality-independent duration-based chunking mechanisms.