Study Reveals Brain Simultaneously Encodes Two Speech Streams During Attention Switching
Key Takeaways
- ▸The brain can simultaneously track two speech streams during attention switching, with new stream encoding beginning before previous stream disengagement
- ▸Alpha power reduction in EEG signals correlates with cognitive effort during different phases of attention reallocation
- ▸Large Language Models were successfully used to model lexical prediction and context-accumulation strategies in human auditory processing
Summary
A landmark neuroscience study published in PLoS Biology reveals that the human brain can simultaneously encode two competing speech streams during attention switching, challenging previous assumptions about how auditory attention works. Using EEG recordings from normal-hearing adults, researchers from Trinity College Dublin and collaborating institutions measured neural encoding patterns while participants switched attention between speech streams in a multi-talker environment every 15–30 seconds.
The research employed Large Language Models to construct context-accumulation strategies and validate lexical prediction mechanisms in the brain. A key finding shows that during attention switches, the neural tracking of a new target stream emerges before disengaging from the previous target, creating a transient period of simultaneous encoding. This transition was accompanied by a reduction in EEG alpha power, indicating increased cognitive effort during attention switching phases. The study also reveals that listeners may reset their lexical context after switching attention, offering fresh insights into how the brain manages dynamic speech processing in complex listening environments.
- The study reveals asymmetric processes during attention switches and suggests the brain resets lexical context when redirecting auditory focus
Editorial Opinion
This research demonstrates the growing value of LLMs as analytical tools for understanding human neuroscience rather than as standalone AI products. By validating how the brain encodes speech against LLM-generated context models, the study illuminates fundamental cognitive mechanisms and provides a new benchmark for evaluating whether AI language models capture human linguistic processing. Such interdisciplinary applications could deepen our understanding of how human and artificial intelligence handle language.


