It is a common practice in language neuroscience to describe speech stimuli as containing phonemes and words. But the ability to infer these units requires knowledge of a type of writing. That there are no acoustic markers in speech serving to divide phonemes and words underlies a long-standing Mapping Problem between units of linguistic description and units identified by neurophysiological methods. Yet research has shown that, in speech listening, low-frequency oscillations in sensory cortices entrain to syllable patterns and groups, indicating that temporal frames of neural processing align with identifiable units in speech sounds. These observations bring to light a basic mismatch between units of linguistic description that link to writing and units of speech processing linked to neurobiological systems. From this viewpoint, the present paper reviews a body of cross-disciplinary findings that provide converging support for a processing of speech in syllable patterns of motion units and sensorimotor chunks. The review focuses on 1) the defining attributes of these units in speech, 2) how they mismatch linguistic concepts, 3) their structural effects across languages, and 4) the behavioral and neurophysiological observations that support their function as processing units. The discussion highlights research on neural entrainment and how observations of auditory-motor coupling in entrained oscillations reveal syllable-related motion units and chunks as sensorimotor units of speech processing. The review also illustrates how these units can address basic issues in linking research on auditory perception and semantic memory to syllables and chunks in utterances as opposed to conceptual units of writing. • Cross-disciplinary reports support syllables and chunks as universal processing units. • Oscillations in sensory cortices align to syllables and chunks as sensorimotor units. • Coupling across auditory and motor cortices attests to sensorimotor units of speech. • Speech processing units address mapping issues in models based on phonemes and words.
Victor J. Boucher (Thu,) studied this question.