Statistical Learning of Prosodic Patterns and Reversal of Perceptual Cues for Sentence Prominence

Recent work has proposed that prominence perception in speech could be driven by predictability of prosodic patterns, connecting prominence perception to the concept of statistical learning. In the present study, we tested the predictability hypothesis by conducting a listening test where subjects were first exposed to a 5-minute stream of sentences with a certain proportion of sentence-final words having either a falling or rising pitch trajectory. After the exposure stage, subjects were asked to grade prominence in a set of novel sentences with similar pitch patterns. The results show that the subjects were significantly more likely to perceive words with low-probability pitch trajectories as prominent independently of the direction of the pitch change. This suggests that even short exposure to prosodic patterns with a certain statistical structure can induce changes in prominence perception, supporting the connection between prominence perception and attentional orientation towards low-probability events in an otherwise predictable context.