Modality-general semantic decoding beyond visual word form

Establish whether EEG decoders trained on silent reading can transfer to passive listening, or whether jointly trained reading-and-listening models contain a shared semantic component that is independent of stimulus-induced visual word-form processing.

Background

Silent-reading EEG decoding may reflect visual and orthographic word-form processing rather than semantic language processing. The paper randomises typography and removes occipital and posterior-temporal electrodes, but explicitly states that neither intervention decisively eliminates the visual confound.

Passive listening is proposed as the critical comparison because auditory language contains no visual word form. Successful transfer from reading to listening, or a shared component in a jointly trained decoder, would provide evidence for modality-general semantic information. The present report does not perform this comparison.