Differential modulation of the neurocomputational signature of early and late mismatch responses to spoken words by awareness and task relevance

Mismatch responses (MMRs), including the mismatch negativity (MMN) and the P3, are a hallmark of the brain’s capacity to detect unexpected events. They are discussed to be generated by two main processes: adaptation and prediction error (PE). However, MMR research on underlying mechanisms has mostly relied on simple, non-linguistic stimuli such as tones, and it remains unknown how awareness and attentional conditions affect MMRs to auditory verbal stimuli, that, unlike tones, carry lexical-semantic meaning. Here, we conducted a preregistered EEG study that disentangled influences of awareness, general task-relevance of word stimuli, and specific target-relevance of word oddballs on the MMN and P3. In addition, single-trial neurocomputational models were applied to uncover the underlying generative models of the MMRs. Results show that word-related MMN and P3 were only elicited when words were task-relevant, and the P3 increased with target-relevance. Crucially, no word-related MMN emerged under inattention, in contrast to the pre-attentive MMN typically reported for tonal deviants. Neurocomputational modeling revealed that the early part of the task-relevant MMN reflects adaptation processes. PE processes generated later parts of the MMN under task-relevance, and the entire MMN under target-relevance, as well as the P3 independent of relevance conditions. This suggests that the same temporal generation pattern of adaptation and PE underlie mismatch responses to both words and tones, but that, unlike tonal MMRs, eliciting MMRs to words requires the allocation of attention. This attentional dependency may reflect the additional lexical-semantic and multi-level representational demands that spoken words impose on the auditory system.