Early vs. Late Planning Hypothesis

(A) Early Planning Hypothesis. The speaker starts planning speech production as soon as enough information is available (B) Late Planning Hypothesis. The speaker defers planning speech production until near the end to avoid overlap

Experimental setup and analysis design

(A) The conversation is segmented into inter-pausal units (IPUs), defined as stretches of continuous speech produced by a single speaker and bounded by silent pauses of at least 200 ms. IPUs serve as the basic units for defining turn transitions and behavioural measures (response latency and duration). (B) For each analysis, participants are designated as either self (yellow) or other (blue). Self refers to the participant whose EEG data are currently being analysed, whereas other refers to their conversational partner. Because analyses are performed symmetrically for both members of the dyad, each participant serves as self during their own listening periods and as other during their partner’s. (C) EEG analyses focus exclusively on the listening interval preceding the self-participant’s speech onset, thereby isolating preparatory neural activity before articulation.

Distributions of behavioural measures across all participants

(A) Response duration, defined as the length of the speaker’s IPU. Only trials included in subsequent EEG analyses are shown (dark gray), corresponding to turn-taking events meeting the latency inclusion criteria. Vertical dashed lines mark the median used for condition splits. (B) Response latency, defined as the time from the offset of the partner’s IPU to the onset of the speaker’s IPU. Negative values indicate temporal overlap between speakers. Dark gray bars indicate trials included in subsequent EEG analyses, whereas light gray bars indicate excluded trials.

Event-related potentials (ERPs) by condition in Fz (top) and Pz (bottom) electrodes.

(Left) ERPs time-locked to response onset for long versus short upcoming IPU durations (inter-pausal units). (Right) ERPs time-locked to response onset for fast versus slow response latencies. Shaded areas indicate the 95% confidence interval across participants (N = 32).

Summary of ERP effects of response duration and response latency.

Topographies of t-values for the duration (top) and latency (bottom) contrasts, time-locked to speech onset. Warm and cool colours indicate relative positivity or negativity for the first condition in each contrast (long - short; fast - slow). Black circles mark electrodes belonging to significant spatio-temporal clusters identified by cluster-based permutation tests (p < 0.05).

Summary of Mixed Effects Model Results

Each row reports the effect of a neural predictor extracted during the listening interval before self speech onset. Neural predictors comprised ERP amplitude, alpha-band activity, and beta-band activity, each summarised within a region of interest (ROI: anterior or posterior) and a pre-speech time window (TW1: −1400 to −800 ms; TW2: −800 to 0 ms). Separate models were fitted for the two behavioural outcomes: upcoming response duration and upcoming response latency. β indicates the standardised fixed-effect coefficient for the neural predictor, SE its standard error, and χ² the likelihood-ratio test statistic obtained by comparing the full model including the neural predictor with a corresponding null model without that predictor. FDR p denotes the false-discovery-rate-corrected p-value across all neural tests within each outcome analysis. The sig. column indicates whether the neural effect survived FDR correction at α = 0.05. All models included participant as a random intercept and controlled for partner duration (other-duration) and the reciprocal behavioural variable (latency when predicting duration; duration when predicting latency)

Alpha/beta power modulation by response duration.

Topographies of t-values for the duration contrast in alpha (top) and beta (bottom) bands, time-locked to speech onset. Warm and cool colours indicate relative positivity or negativity (long - short). Black circles mark electrodes belonging to significant spatio-temporal clusters identified by cluster-based permutation tests (p < 0.05).

Temporal generalisation decoding of response duration and latency.

(A) Diagonal decoding accuracy (area under the ROC curve, AUC) for classifying short vs. long responses (left) and fast vs. slow responses (right). Shaded areas indicate 95% confidence interval across participants; The black outlines mark the statistically significant clusters (p < 0.05). (B) Temporal generalisation matrices showing decoding performance across training (y-axis) and testing (x-axis) times relative to speech onset (0 s). Red regions indicate above-chance classification and black contours denote significant clusters. Both duration and latency could be decoded from neural activity beginning ∼1.3 s before speech onset, with large, temporally stable patterns extending up to articulation.

Conceptual model of early maintenance and late commitment during turn preparation

The figure is schematic and intended to illustrate a possible functional organisation consistent with the observed neural dynamics. (A) Speech planning unfolds in two stages: an early content activation stage followed by a later motor-preparation stage. (B) With more content prepared, multiple candidate contents are co-activated and maintained in the background. This is associated with higher alpha power (top) and a larger set of active neural assemblies (bottom). Green assemblies denote production-related processes and orange assemblies denote comprehension-related processes; clusters represent distinct cortical areas. At this stage, only comprehension-related assemblies are bound across areas. (C) With less content prepared, fewer background assemblies are active and alpha power is lower. (D) During late motor preparation, production-related processes become more strongly coordinated as speech onset approaches, while comprehension-related processes recede to the background.

Spatiotemporal distribution of ERP effects for duration and latency conditions.

Topographic maps show the scalp distribution of ERP t-values computed for the contrasts of interest. Maps are displayed in 100 ms time steps relative to the event of interest. Panel (A) shows the effect of response duration, and Panel (B) shows the effect of response latency. Warmer colours indicate more positive t-values and cooler colours indicate more negative t-values. Each map represents the spatial distribution of the statistical effect across electrodes within the corresponding time window.

Spatiotemporal distribution of alpha power effects for duration and latency conditions.

Topographic maps show the scalp distribution of t-values for alpha-band power computed for the contrasts of interest. Maps are displayed in 100 ms time steps relative to the event of interest. Panel (A) shows the effect of stimulus duration, and Panel (B) shows the effect of response latency. Warmer colors indicate more positive t-values and cooler colors indicate more negative t-values. Each map represents the spatial distribution of the statistical effect across electrodes within the corresponding time window.

Spatiotemporal distribution of beta power effects for duration and latency conditions.

Topographic maps show the scalp distribution of t-values for beta-band power computed for the contrasts of interest. Maps are displayed in 100 ms time steps relative to the event of interest. Panel (A) shows the effect of stimulus duration, and Panel (B) shows the effect of response latency. Warmer colors indicate more positive t-values and cooler colors indicate more negative t-values. Each map represents the spatial distribution of the statistical effect across electrodes within the corresponding time window.

Joint ERP plots by speech duration condition.

Joint plots showing the spatiotemporal dynamics of the event-related potential (ERP) for the two speech-duration conditions. (A) Long-duration condition. (B) Short-duration condition. In each panel, the ERP time course is displayed together with scalp topographies at representative latencies, illustrating the spatial distribution of the evoked activity associated with each duration condition.

Joint ERP plots by response latency condition.

Joint plots showing the spatiotemporal dynamics of the event-related potential (ERP) for the two response-latency conditions. (A) Fast response condition. (B) Slow response condition. In each panel, the time course of the ERP is shown together with scalp topographies at representative latencies, illustrating the spatial distribution of the evoked activity underlying the waveform for each condition.

Relationship between ERP morphology and response latency distribution at frontal and parietal electrodes.

Event-related potentials (ERPs) were computed after splitting trials by median response latency. Dark red traces indicate trials with faster responses, and light red traces indicate trials with slower responses; shaded areas represent the standard error of the mean. ERPs are shown for electrodes Fz (left) and Pz (right). Histograms below each ERP panel display the distribution of response latencies for the same trials used in the median split. At Fz, the temporal profile of the ERP closely mirrors the latency distribution: the main inflection in the ERP occurs near the peak of the response-latency histogram. This correspondence suggests that the apparent ERP modulation largely reflects alignment to auditory offset timing rather than a genuine latency-related neural difference. In contrast, at Pz the ERP waveform does not follow the latency distribution as closely, indicating that the observed ERP differences more likely reflect a true response-latency-related neural effect rather than a simple consequence of the underlying latency distribution.