Unlocking Brain's Memory with Language Models

Have you ever wondered how your brain remembers personal and famous people and places? Scientists are exploring this by turning memory into semantic vectors using language models. They asked participants to name their closest people and places, then asked questions about these memories. They recorded brain signals while participants read these names and imagined the related people or places. For familiar entities, they used the participants' answers, and for famous ones, they used Wikipedia pages. By analyzing brain waves, they found that language models could predict how the brain processes these memories. Early responses were better predicted in the temporo-parietal region, while later responses were better in the frontal and central regions. A more advanced language model, XLM, outperformed a simpler one like word2vec. This shows how language models can decode subject-specific memories in the brain.