In the eyes of a language model: A comprehensive examination through eye-tracking data

Published: 01 Jan 2025, Last Modified: 09 Oct 2025Neurocomputing 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Multilingual interpretability study of NLMs informed by human reading behavior.•Eye-tracking fine-tuning strengthens links between attention and human gaze.•Intermediate ET fine-tuning does not degrade model performance on downstream tasks.
Loading