"When the word ends, the story begins": word offsets improve encoding of natural speech in a MEG study

The segmentation of continuous speech into words was investigated using the temporal response function (TRF) approach for source-space magnetoencephalography (MEG) signals from 54 participants. Although the onset of words has been extensively studied, the contribution of their offset remains unclear. Acoustic and semantic predictors were used to test the relative contribution of word onsets and offsets to the processing of natural speech in the auditory temporal cortex. Circular shifting of word onsets and offsets significantly reduced predictive power (p < 0.001 for both). The corrupted word onset model showed a greater reduction in the predictive performance (p < 0.001). Both effects were localised to the left middle temporal gyrus (MTG) and the right banks of the superior temporal sulcus (STS). Additionally, removing the word offset predictor from the model significantly reduced predictive performance, indicating incremental word offset timing information beyond word onset and the remaining control predictors. These findings demonstrate the significant, albeit unequal, contribution of word onsets and offsets to the neural encoding of natural speech.