Can Brain Signals Reveal Inner Alignment with Human Languages?

Code for our paper Can Brain Signals Reveal Inner Alignment with Human Languages?.

In EMNLP Findings 2023.

Usage

Set up the Environment

Create a virtual environment and activate it.

python -m venv .env

source .env/bin/activate

Install basic requirements.

pip install -r requirements.txt

Download Datasets

Download K-EmoCon Dataset here.

Download ZuCo Dataset here.

For ZuCo Dataset, please only download task1 and task3.

Preprocessing data

In the preprocessed folder, preprocessed data is readily available for usage. For K-EmoCon, df.csv is used. For ZuCo sentiment analysis, we provide the sentence-level csv in the preprocessed folder.

Preprocessing scripts are provided as well.

Training

The main_new.py file is used for training selected models. Arguments are provided for selecting datasets, modalities, models, levels, and tasks. Please view the config.py file in tandem and customize it as necessary.

Plotting

The plot.py file is used for plotting TSNE, alignment, and brain topological figures. Arguments are also provided for selecting datasets, modalities, models, levels, tasks, and types of plots. Additionally, plot_new.py is used to plot the learning curves.

Citation

@misc{han2023brain,
      title={Can Brain Signals Reveal Inner Alignment with Human Languages?}, 
      author={William Han and Jielin Qiu and Jiacheng Zhu and Mengdi Xu and Douglas Weber and Bo Li and Ding Zhao},
      year={2023},
      eprint={2208.06348},
      archivePrefix={arXiv},
      primaryClass={q-bio.NC}
}

Contact

If you have any questions, please contact wjhan@andrew.cmu.edu, jielinq@andrew.cmu.edu.