Benchmarking Sentence Processing Models: Do Surprisal and Lossy-Context Surprisal Outperform Theory?

Michael Vrazitulis, Pia Schoknecht, Shravan Vasishth

Published: 03 Oct 2025, Last Modified: 03 Oct 2025CPL 2025 PosterEveryoneRevisionsBibTeXCC BY 4.0
Keywords: sentence processing, surprisal, lossy-context surprisal, self-paced reading, eye tracking, model comparison, PSIS-LOO
TL;DR: We compare surprisal, lossy-context surprisal, and theory-based qualitative predicions of reading times using GEPPU, a novel benchmark dataset for German sentence processing.
Submission Number: 43