Chained melody: Low-level acoustic cues as a guide to hierarchical structure in comprehension.

Trotter, A. S., Monaghan, P., & Frost, R. L. A. (2017). Chained melody: low level acoustic cues as a guide to hierarchical structure in comprehension. Poster presented at Interdisciplinary Advances in Statistical Learning, Bilbao, Spain. 

Abstract:

To accurately process and respond to speech requires rapidly unpacking its structural dependencies as well as comprehension of meaning. Theoretical approaches are divided as to whether syntactic processing requires hierarchical phrase structures or lower-level statistical correspondences. In this study, we investigated the extent to which prosody – the rhythmic and melodic aspects of speech – may support low-level processing of complex syntactic structures. We hypothesised that elements of a sentence that contain syntactic dependencies would be similar in terms of pitch, enabling grouping according to the gestalt similarity principle.

We analysed data from American English speakers (n = 64) spontaneously producing either passive (e.g., [the bear]1 [held]2 [by the girl]3 [is green]4) or hierarchical centre-embedded (HCEs, e.g., [the bear]1 [the girl]2 [held]3 [is green]4) sentences elicited using a picture description task, and divided them into four phrase positions (see indices in examples). Using linear mixed effects modelling, we found a smaller pitch decrease from first to second positions in passives, and a smaller pitch decrease from second to third positions in HCEs. These different pitch progressions enable the similarity gestalt to support syntactic dependencies and deter interpretations consistent with canonical word order in passives, consistent with lowlevel computations driving syntactic processing.