EEG data on indirect request processing

This dataset contains EEG recordings from 21 Korean native speakers (seven males, fourteen females, zero other response; age M = 25.8, SD = 1.9) while they read dialogues containing literal responses or indirect requests. [manuscript abtract]Pragmatic language comprehension, such as indirect request processing, requires going beyond the literal meaning by integrating contextual and linguistic cues. Although event-related potential (ERP) studies have identified neural markers such as P600, they offer limited insight into EEG’s complex temporal and spatial dynamics. To address these limitations, this study applies neural decoding to EEG data by leveraging a convolutional transformer model (EEG-Conformer; Song et al., 2023) to classify neural responses to indirect requests versus literal responses. Whereas the ERP analysis revealed significant effects at 500-800 ms for indirect requests but no effects at an earlier time window, the neural decoding results showed peak classification accuracy (70.0%) at 300-500 ms, 400-600 ms and 500-700 ms time windows, suggesting early differentiation for pragmatic processing. Further analysis demonstrated that replacing the CNN module in EEG-Conformer with AlexNet, LeNet-5, or ResNet improved performance, with ResNet achieving the highest accuracy (90.0%) at 500-700ms. These findings highlight the potential of deep learning-based EEG analysis in neuropragmatics, providing a new approach for investigating pragmatic inference and intention processing beyond traditional ERP methods.