Temporal grouping and sequence inference drive pattern detection in fast multi-feature sequences

As we experience the world around us, we naturally track the statistical predictability of unfolding auditory sequences and form expectations about upcoming events. Typically, the statistical properties of multiple acoustic dimensions (or features) co-vary forming predictable auditory patterns. Yet most research has focused on how the predictability of individual features affects auditory pattern detection, with few having examined how multiple co-varying features affect each other’s predictability. Here, stimuli were rapid tone sequences with statistical predictability varying in two dimensions: a first “ignored” feature (tone frequency) was either random or structured in a regularly repeating pattern, while a second “target” feature (timing, loudness, or location) contained transitions from a regular to random order. Participants were instructed to ignore the non-informative frequency feature and focus on detecting transitions in the target feature. When the regularity cycles of the ignored feature matched the cycle of the target feature, there was a large effect of frequency predictability on change detection in the target feature, demonstrating a large benefit of co-varying predictability. Importantly, a small but still significant effect was observed even when the cycles were mismatched indicating that listeners automatically track predictability within independently varying stimulus features. Together, our results elucidate perceptual effects which determine how acoustic features interact when forming expectations.