Inferring Native and Non-Native Human Reading Comprehension and Subjective Text Difficulty from Scanpaths in Reading
This repo provides the code for reproducing the experiments in Inferring Native and Non-Native Human Reading Comprehension and Subjective Text Difficulty from Scanpaths in Reading.
The figure above shows our architecture
BEyeLSTM
.
We investigate and show that we can generalize to unseen test persons for all tasks investigated.
You can clone this repository by either using
git clone git@github.com:aeye-lab/etra-reading-comprehension
cd etra-reading-comprehension
or
git clone https://github.com/aeye-lab/etra-reading-comprehension
cd etra-reading-comprehension
depending on your preferences and settings.
Afterward, change into the directory by using cd etra-reading-comprehension
.
You can download the publicly available data here
git clone git@github.com:ahnchive/SB-SAT
or
git clone https://github.com/ahnchive/SB-SAT
Install all required python packages via:
pip install -r requirements.txt
You can create the data splits using:
python3 utils/generate_text_sequence_splits.py
Then you can directly start using both BEyeLSTM and the baseline of Ahn et al. using python3 nn/train_model.py
or python3 ahn_baseline/evaluate_ahn_baseline.py
respectively. By changing the boolean arguments in nn/train_model.py
you can recreate our ablation study or use different subnets only.
Note: Running the experiments, especially on CPU, will take some time.
If you find any issues, please open an issue in the issue tracker.
If you use our code for your research, please consider citing our paper:
@inproceedings{10.1145/3517031.3529639,
author = {Reich, David Robert and Prasse, Paul and Tschirner, Chiara and Haller, Patrick and Goldhammer, Frank and J\"{a}ger, Lena A.},
title = {Inferring Native and Non-Native Human Reading Comprehension and Subjective Text Difficulty from Scanpaths in Reading},
year = {2022},
isbn = {9781450392525},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
doi = {10.1145/3517031.3529639},
booktitle = {2022 Symposium on Eye Tracking Research and Applications},
articleno = {23},
numpages = {8},
keywords = {deep learning, reading comprehension, eye tracking-while-reading},
location = {Seattle, WA, USA},
series = {ETRA '22}
}