Frequency Explains the Inverse Correlation of Large Language Models' Size, Training Data Amount, and Surprisal's Fit to Reading Times.

Byung-Doh Oh, Shisen Yue, William Schuler

Browse the full EACL paper archive.