Protracted development in children of perceptual segregation of competing talking faces in the multisensory cocktail party problem

Successful communication with our social partners requires binding, integrating, and perceptually segregating the audible and visible attributes of the multiple talking faces that we often encounter in social situations, a challenge known as the multisensory cocktail party problem (MCPP). Although audiovisual (AV) temporal synchrony is a powerful cue for binding speech signals, how children develop the ability to use this cue to segregate a target talker remains unclear. Here, we examined the development of gaze dynamics supporting multisensory segregation in 3-7-year-old children (N = 149) and adults (N = 37) viewing four talking faces accompanied by a single auditory utterance synchronized with one of the faces (i.e., target). Using metrics of gaze dynamics from information theory, namely proportion of total looking time, stationary entropy, transition entropy, and transition rates, we show that even though sensitivity to AV synchrony is present by age 3, it is insufficient for efficient target segregation. It is not until ages 5-6, following a qualitative shift in dynamic gaze control and more structured distractor transitions, that target selection becomes more efficient, but still not as efficient as it is in adults. We interpret these developmental changes as reflecting a shift from early detection of multisensory cues to later-emerging strategies that organize visual sampling in relation to auditory information in a task-dependent manner. Together, they demonstrate that solving complex multisensory challenges depends on AV integration as well as on the development of dynamic gaze organization that supports efficient multisensory perceptual segregation over time. Significance StatementSocial communication requires segregating one talker from others, a challenge known as the multisensory cocktail party problem. Although adults solve this efficiently, how this ability develops remains unclear. Using dynamic gaze measures derived from information theory, we show that multisensory segregation in childhood depends not only on detecting audiovisual synchrony but also on the emergence of structured gaze strategies. Only by ages 5-6 do children combine sustained target fixation with organized sampling of competing talkers. Even by age 7, these audiovisually guided strategies remain immature relative to adults. These findings reveal probabilistic sampling mechanisms through which gaze supports multisensory segregation, offering a mechanistic account of how children learn to navigate complex social environments, with implications for language development and education.