Lisha Li

The typical approach to decoding speech from the brain (using brain-machine interfaces) is to decode low-level linguistic units (e.g. phonemes, syllables) from motor articulation areas (e.g. Premotor cortex) with the aim of assembling these low-level units into higher-level discourse. We propose that brain-to-speech decoding may benefit from adopting a functional view of language, which conceives of language as an instrumental tool for interacting with others' intentions in order to fulfil one's own intentions. This functional view of language motivates adopting usability (i.e. the decoder's usefulness as a tool for achieving goals), in addition to decoding accuracy, as a criterion for assessing decoder performance. Decoders may achieve gains in usability by incorporating data about communicative situations and speaker intentions in order to generate and fill in speech act templates. We suggest that this intention-oriented, template-based, and functionally inspired view of brain-to-speech decoding may facilitate efforts to achieve naturalistic speech decoding.