Back

What are you talking about? Representation of the Topic of Speech in the Human Brain

Boncz, A.; Toth, B.; Winkler, I.

2024-12-12 neuroscience
10.1101/2024.12.12.628085 bioRxiv
Show abstract

A sentence outside context provides very restricted information to the listener. Real understanding requires one to be able to place it within the sequence of sentences forming a coherent "story" and just as importantly, within ones general knowledge of the "topic". Here we demonstrate the existence of functional brain networks robustly sensitive to the topic of ca. 6-minute-long coherent newspaper articles. Main network hubs are located in parietal and temporal brain regions. These networks changed very slowly during the course of the article, consistent with their involvement in representing the topic rather than the story aspect of the context. Because even the meaning of words can depend on the topic (see e.g., the word "shuttle" in transportation and weaving) the topic-sensitive network must also be intimately linked with the mental lexicon underlying language comprehension. Significance StatementWhen we listen to someone talking, our brain continuously integrates the speech stream with our general knowledge associated with the topic. While this is a central process of comprehension, it is still unclear how the brain maintains the relevant representations. Here we demonstrate the existence of a cortical network, which is sensitive to the topic of the speech just heard. This network may form the neural substrate for "topic representation" in the brain. Understanding how such representations are maintained and accessed would let us reveal the computations enabling rapid linguistic integration, a key process in human communication.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.