Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Contextual Prediction Tunes the Tempo of Speech Segmentation
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Olesia Platonova1, Olesia Dogonasheva1, Anne-Lise Giraud1,2, Sophie Bouton1,3; 1Université Paris Cité, Institut Pasteur, AP-HP, INSERM, CNRS, Fondation Pour l'Audition, Institut de l'Audition, IHU reConnect, F-75012 Paris, France, 2Department of Basic Neuroscience, University of Geneva, 3Laboratoire de Sciences Cognitives et Psycholinguistique, Ecole Normale Supérieure, Université PSL, CNRS, Paris, France
Speech comprehension relies on both temporal structure and contextual prediction. Yet, it is poorly understood how these mechanisms coordinate to assure stable comprehension even under rapid replacement of units unfolding in time. Oscillatory accounts propose that this may be the role of brainwaves, with flexible theta-band oscillations (~4-8 Hz) supporting temporal parsing of speech, while beta-band oscillations (~12-30 Hz) are implicated in maintaining and updating contextual predictions. In the present study, we use time-compression to degrade temporal structure while preserving linguistic content and investigate how information delivery rate, temporal regularity, and linguistic structure jointly shape speech comprehension. Across two behavioral ‘re-packaging’ experiments (N = 110 neurotypical adults), x3 time-compressed sentences are combined with periodic insertions of silences between the speech units to manipulate (i) delivery rate of sound packages, (ii) rhythmic regularity of such delivery, and (iii) boundary alignment (syllable-based vs. strict duration time-based segmentation). In addition, we quantify contextual predictability using word-level entropy to assess the contribution of top-down constraints. Results reveal (i) a non-linear effect of delivery rate: comprehension peaks at rates exceeding the canonical theta range (~8–13 Hz) and declines at slower or faster pacing (p < .001 for both Exp 1 and 2). (ii) Temporal regularity benefits word recognition only when boundaries coincide with syllabic units, indicating that periodic pacing alone is insufficient for effective parsing. (iii) Syllable-based alignment of packages significantly improves word recognition relative to time-based segmentation (p < .001), with strongest benefits outside the optimal rate range (e.g., 10.8 Hz: p < .001). Crucially, contextual predictability (word-level entropy) facilitates comprehension primarily when temporal cues are least supportive: outside the optimal theta range under syllable-aligned segmentation (three-way interaction Exp.1: p = .048; Exp.2: p = .031). Theta mediation is efficient for comprehension when the speech is delivered within its optimal regime, while fades outside this regime, allowing the active role of beta-mediated predictive processes. Together, these results suggest that contextual prediction is continuously active but behaviorally visible only when temporal scaffolding is insufficient and syllabic structure is preserved.
Topic Areas: Speech Perception, Speech-Language Treatment