Gaze-guided sentence simplification

This repository contains the code and data for reproducing the results in the paper "Exploring Cognitively Informed Sentence Simplification with Gaze-Guided Text Generation" (Gaze4NLP 2026).

Reproducing results

Requires Python>=3.12.

Setup

pip install -r requirements.txt
pip install --no-deps lens-metric==0.2.0

Training gaze models

bash train_gaze_models.sh

This trains all gaze model configurations and saves them in gaze-models/.

Scoring texts

bash score_texts.sh

This predicts gaze scores for the reference simplifications in the ASSET validation set and saves them in gaze-scores/.

Generating simplified texts

bash generate_texts.sh

This generates outputs for the ASSET validation set with all gaze model configurations and saves them in outputs/.

Evaluating simplified texts

bash evaluate_texts.sh

This calculates evaluation metrics for all outputs and saves them in results/.

Generating figures

python analysis.py

This generates the figures in the paper and saves them in results/.

Licenses

Eye-tracking data (gaze-data/): The EMTeC, OneStop, MECO, and Provo corpora are all licensed under CC-BY-4.0.

Sentence simplification data (simplification-data): The ASSET dataset is licensed under CC-BY-NC-4.0.

The code in this repository is licensed under MIT.