A disassociated neural network underlies the dynamic construction of speech mental imagery independent of auditory perception.

Prominent transitions between different mental contexts produce a stereotyped brain state shared across internally and externally generated transitions but distinct from minor within-context transitions, potentially reflecting a major flushing and updating of mental models.

The silent production of words in one's mind generates an efference copy that is similar in nature to the efference copy associated with overt vocalization.

Remaining focused on the topic at hand when speaking depends on effective selection of task-relevant semantic knowledge, and declines in this ability account for increases in off-topic speech in older people.

Neuroimaging reveals that the brain response to spoken language can be better explained by rule-based models than statistical models recently developed in artificial intelligence research.

Structural representations of sentences generated by deep language models correlate with human listeners' behaviours and neural activity, providing a quantifiable framework to uncover the neural dynamics underpinning incremental speech comprehension.

In the processing of spoken narratives, bottom-up acoustic cues and top-down linguistic knowledge separately contribute to neural construction of linguistic units.

Full-text analysis of the human brain-imaging literature can predict the spatial distribution of observations in the brain given the description of a subject of study.

Hidden Markov Modeling to fMRI data reveals how language comprehension arises from the dynamic interplay of large-scale brain networks.

A non-invasive cognitive assistant for blind people endows objects in the environment with voices, allowing users to explore the scene, localize objects, and navigate through a building with minimal training.