Dependency, Structure & Memory: A Reading Time Benchmark for Sentence Processing Model Evaluation

Nina Nusbaumer, Corentin Bel, Iria de-Dios-Flores, Guillaume Wisniewski, Benoit Crabbé

Published: 03 Oct 2025, Last Modified: 03 Oct 2025CPL 2025 SpotlightPosterEveryoneRevisionsBibTeXCC BY 4.0
Keywords: sentence processing, working memory, benchmark, dataset, model evaluation
TL;DR: A new reading-time dataset systematically varies syntactic complexity and dependency length, includes working memory scores for 540 participants, and offers a controlled, cognitively grounded benchmark for syntax-sensitive language model evaluation.
Submission Number: 44