arXiv - CSCL: "Transformer-Based LM Surprisal Predicts Human Rea…" - Qoto Mastodon

arXiv - CSCL @arxiv_cscl@qoto.org

Transformer-Based LM Surprisal Predicts Human Reading Times Best with About Two Billion Training Tokens. (arXiv:2304.11389v1 [cs.CL])

http://arxiv.org/abs/2304.11389 #arXiv #NLProc

Apr 25, 2023, 03:06 · · arxiv-cscl · · ·

Sign in to participate in the conversation