Does prosodic entrainment influence segmentation? Towards a predictive account of prosody

Prosody plays an important role in sentence segmentation. Cues such as pauses, pitch modulations, and duration changes trigger the bottom-up segmentation of sentences into syntactic constituents (Frazier et al., 2006). Earlier priming studies found that prior prosodic cues can also implicitly affect the segmentation of downstream sentences, even if these lack prosody themselves (Steinhauer & Friederici, 2001). Recently, the processing of prosody has been linked to cyclic electrophysiological activity, so-called neural oscillations. In particular, delta-band oscillations (i.e., < 4 Hz) synchronize or entrain to the rhythm of intonational phrase boundaries (IPBs; for review, see Meyer, 2018). Via entrainment, oscillations inherit acoustic rhythms to persist after stimulation offset (e.g., Kösem et al., 2018). We here hypothesized that entrainment could induce implicit prosody and thereby affect the segmentation of a downstream sentence. Specifically, we hypothesized that auditory entrainment would trigger the perception of an IPB in an upcoming visual sentence, leading to the termination of a syntactic constituent in spite of the absence of an IPB. Two experiments were run on 80 participants each via the Gorilla framework. Trials started with one of two prosodic contours, repeated three times to induce entrainment. The contour was followed by word-by-word RSVP of a target sentence; the sentence-final words were self-paced. We aimed to affect segmentation of the target by the preceding contour. Segmentation was inferred from reading times to the self-paced words, as well as from reaction times to correctly answered comprehension questions. Statistical analysis was based on systematic comparison of mixed linear models. In the first experiment, target 2-clause sentences involved a coordination ambiguity such as Jim saw Tom and John laughed (Hoeks et al., 2002).Target 1-clause sentences (e.g., Jim saw Tom and John.) were used to avoid habituation. Of the two types of prosodic contours, the slow contour matched the duration of Jim saw Tom and John, aimed at driving participants up the garden path in 2-clause sentences. In contrast, the fast contour matched the duration of Jim saw Tom, aimed at avoiding the garden-path. For 1-clause sentences, correct syntactic structures were expected in case of slow entrainment, but incorrect ones in case of fast entrainment. Reading times did not show the hypothesized garden-path effect at the verb. Instead, readers were found to selectively speed up on the third noun under slow entrainment for both 1– and 2-clause sentences. We interpreted this pattern as evidence that slow entrainment strengthened participants’ temporal prediction of a 2-clause (i.e., long) sentence, as the participants were speeding up in the expectation of a final verb. Prosodic entrainment thus appears to affect segmentation of upcoming sentences through temporal prediction. Results from a second experiment strengthen the temporal prediction account. The experiment used non-ambiguous target sentences: two-NP (e.g., Max sees Tom and Karl) and one-NP (e.g., Max sees Tom). We replicated the self-paced reading effect: a speed-up under slow entrainment. In addition, we obtained a cross-over interaction in comprehension questions: slow contours induced faster reaction times for two-NP sentences, and slower reaction times for one-NP sentences, while fast contours had an opposite effect. Our results suggest that prosody entrainment may serve a predictive function at the prosody–syntax interface: Via neural entrainment, prosodic contours may enable the prediction of the duration of upcoming sentences or segments. Entrainment may thus explain why prosody is not limited to bottom-up segmentation, but extends to top-down segmentation (Breen, 2014; Frazier et al., 2006; Grosjean, 1983)