Using large cortical recording coverage with ECoG, previous influential studies showed a distributed representation of phonemes over the superior temporal lobe, which appeared to follow a feature-based logic. Such a widespread, mesoscopic-scale, representational geometry of phonemes is intriguing, yet conflicts with the classical notion of hierarchical processing. Here we assumed that early, focal, spectro-temporal representations of speech sounds are sufficiently informative to permit rapid recognition without involving distributed cortical patterns. This hypothesis implies that previous findings reflect redundant integrated neural activity that is not essential for phonological recognition. Using fMRI and MEG during syllable identification, we first show that sensory and decisional activity co-localize to a restricted part of the posterior superior temporal cortex. Next, using intracortical recordings we demonstrate that early and focal neural activity in this region distinguishes correct from incorrect decisions and can be machine-decoded to classify syllables. Crucially, significant machine-decoding was possible from neuronal activity sampled across widespread regions, despite weak or absent sensory or decision-related responses. These findings concur to demonstrate that focal neural tracking of acoustic cues is both necessary and sufficient to account for phonemic categorization and that meso-scale representational geometry is hence marginally relevant for human decision.
Bouton, S., Chambon, V., Tyrand, R., Guggisberg, A. G., Seeck, M., Karkar, S., Van de Ville, D., Giraud, A.-L. (2018). Focal versus distributed temporal cortex activity for speech sound category assignment. PNAS. https://doi.org/https://doi.org/10.1101/133272