Seminar

Restoring communication in patients with anarthria remains a major clinical challenge and requires the development of invasive brain–computer interfaces (BCIs) capable of decoding inner speech directly from neural activity. While current approaches have achieved impressive decoding performance, they primarily target the sensorimotor cortex and are therefore unlikely to benefit many patients with cortical impairments, such as post-stroke aphasia. Our work investigates an alternative strategy based on decoding from distributed language networks, particularly temporo-parietal auditory and somatosensory regions that support phonological representations. We focus on the syllabic level as an intermediate linguistic unit that balances computational tractability with the production of naturalistic speech output. Using both invasive and non-invasive recordings, we examine neural features across the phonological network that enable inner speech decoding and study how patients adapt neurally to improve BCI control. Ultimately, this research aims to support the development of a chronic, wireless, fully implantable speech BCI capable of restoring communication at rates approaching natural language.