The study introduces the Composition Score to quantify meaning composition during sentence comprehension using Large Language Models. It correlates with brain clusters linked to word frequency and structural processing, revealing insights into human language comprehension.
The research explores the process of meaning composition in the human brain using a novel metric called the Composition Score derived from Large Language Models. The study reveals correlations between this metric and brain regions associated with word frequency, structural processing, and general sensitivity to words. By analyzing patterns of high and low scores across different layers of models, the study provides insights into how LLMs optimize memory efficiency. Additionally, regression analyses show that the Composition Score outperforms traditional control variables like word frequency and syntactic node counts in predicting neural activity. The findings suggest a multifaceted nature of meaning composition during human sentence comprehension.
Ke Bahasa Lain
dari konten sumber
arxiv.org
Wawasan Utama Disaring Dari
by Changjiang G... pada arxiv.org 03-08-2024
https://arxiv.org/pdf/2403.04325.pdfPertanyaan yang Lebih Dalam