Extracting language content from speech sounds: An information theoretic approach

Speech comprehension involves recovering the speaker's intended meaning from the speech sounds that they produce. While the sensory-driven components of this process have been widely investigated , the impact of speech content (i.e. linguistic information) on sensory processing is much less understood. Here we summarise the growing body of research demonstrating that neural processing of speech sounds is influenced by morpheme-and word-level statistical properties of the information conveyed. We introduce and review evidence that information theoretic measures such as entropy and surprisal are apparent in neural responses. These findings help uncover fundamental organisational principles of the language system: what units are stored and how they are accessed. Modelling sensitivity to the information content of the speech signal helps explain the interface between (i) auditory processes operating on speech sounds and (ii) the words and meanings that those sounds convey.