Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Decoding the Time Course of Intonation Planning in Speech Production
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Helin Erden1, Yiya Chen2, Antje Meyer1, Constantijn van der Burght1,2; 1Max Planck Institute for Psycholinguistics, 2Leiden University
In speech production, the temporal relationship between sentence-level intonation planning and word-level phonological encoding remains poorly understood. Prosody-first accounts propose that an abstract prosodic frame is established early during planning, before lower-level prosodic elements and segmental content are retrieved. In contrast, prosody-last accounts treat intonation as a late addition, superimposed onto an already-assembled phonological string. Therefore, these frameworks make opposing predictions about the relative neural onset of intonation encoding compared to other levels of linguistic encoding, particularly word-level prosody (lexical stress). The present study used EEG with time-resolved multivariate pattern analysis (MVPA) to empirically address this question. Thirty native Dutch speakers named pictures of disyllabic Dutch nouns using either a rising (Question) or falling (Exclamation) pitch contour, cued by the color of a frame presented around the target picture. Target words were orthogonally balanced across four additional linguistic dimensions targeting distinct production stages: animacy (conceptual-semantic encoding), lexical stress (word-level prosodic encoding), and manner and place of articulation of the word-initial phoneme (phonetic-articulatory encoding). Time-resolved decoding and temporal generalization analyses were conducted using a classifier characterizing the time course of each linguistic dimension during the speech planning period. Because our primary question concerned relative onset differences between dimensions, we further estimated the uncertainty of decoding onset latencies using a divergence-point bootstrap procedure, which enabled statistical comparison of pairwise onset latency differences. Acoustic analyses confirmed that questions and exclamations were produced with rising and falling pitch contours as intended. The main MVPA results revealed a temporal ordering broadly consistent with cascaded models of production: Animacy was the earliest dimension to become decodable (onset: 141 ms; 95% CI: 111-151 ms), followed by intonation (358 ms; CI: 225-465 ms) and place of articulation (468 ms; CI: 231-548 ms) at statistically comparable latencies, with manner of articulation emerging latest (618 ms; CI: 448-781 ms). Lexical stress was not reliably decodable during the speech planning interval, suggesting that the global intonation signal may have masked a separable lexical stress representation, given that both are primarily realized through pitch in Dutch. Further exploratory analyses decoded intonation separately within word-initial and word-final stress subsets. The results revealed that intonation decoding emerged significantly earlier for word-final stress words than word-initial stress words (95% CI of the difference: 3-433 ms), indicating that the temporal profile of intonation planning is shaped by the word’s stress pattern. This asymmetry may be related to differences in the frequencies of the two stress patterns in Dutch. These findings provide the first demonstration that sentence-level intonation is neurally decodable from EEG during overt speech planning. The comparable onset of intonation and phonetic-articulatory decoding argues against an early, independently established prosodic frame, and is more consistent with a prosody-last account. Importantly, the modulation of intonation decoding by lexical stress pattern suggests that sentence-level and word-level prosody closely interact and evolve together during planning, rather than being encoded independently.
Topic Areas: Prosody, Language Production