Toddlers' action prediction: Statistical learning of continuous action sequences

- Statistical learning supports predictive eye movements during observation of action sequences.
- Toddlers predicted more accurately from observing an actor than a visual event sequence.
- Anticipatory eye movement behavior emerges during online learning in young toddlers.
- Observing actions, but not visual events, influenced toddlers' action choices when associated with an effect.

The current eye-tracking study investigated whether toddlers use statistical information to make anticipatory eye movements while observing continuous action sequences. In two conditions, 19-month-old participants watched either a person performing an action sequence (Agent condition) or a self-propelled visual event sequence (Ghost condition). Both sequences featured a statistical structure in which certain action pairs occurred with deterministic transitional probabilities. Toddlers learned the transitional probabilities between the action steps of the deterministic action pairs and made predictive fixations to the location of the next action in the Agent condition but not in the Ghost condition. These findings suggest that young toddlers gain unique information from the statistical structure contained within action sequences and are able to successfully predict upcoming action steps based on this acquired knowledge. Furthermore, predictive gaze behavior was correlated with reproduction of sequential actions following exposure to statistical regularities. This study extends previous developmental work by showing that statistical learning can guide the emergence of anticipatory eye movements during observation of continuous action sequences.