Learning the meanings of function words from grounded language using a visual question answering model
This repository contains all the code for models, probes, and evaluations presented in the paper Learning the meanings of function words from grounded language using a visual question answering model by Eva Portelance, Michael C. Frank, and Dan Jurafsky. 2023.
All the necessary preprocessed data -- the CLEVR dataset (Johnson et al. 2017) partitions and semantic probes -- are available for download at this OSF project repository.
More probe questions and partitions can also be created by using the scripts and notebooks provided in the folder ./dataset and probe creation/.
All pretrained models are available for download at this OneDrive link.
The model code is available in the folder ./model and evaluations/mac-network. This folder contains a custom modified version of the MAC model by Hudson & Manning (2018).
All the evaluation and result analysis scripts are available in the folder ./model and evaluation/probe-evaluation-collection/. The results from the paper are also available for download at this OSF project repository.
Use the following to set up a compatible environment.
- Python 3.7 or higher
- pip package manager
- GPU (for faster training, optional but recommended)
Clone the repository:
git clone https://github.com/evaportelance/vqa-function-word-learning.git
If you use anaconda, you can clone our environment using the conda-env.txt file:
cd vqa-function-word-learning
conda create --name myenv --file ./conda-env.txt
pip install requirements.txt
Please cite the following paper:
@article{portelance2024learning,
title={Learning the meanings of function words from grounded language using a visual question answering model},
author={Portelance, Eva and Frank, Michael C. and Jurafsky, Dan},
year={2024},
journal={Cognitive Science}
}
Please also cite the papers for the MAC model and the CLEVR dataset:
@inproceedings{hudson2018compositional,
title={Compositional Attention Networks for Machine Reasoning},
author={Hudson, Drew A and Manning, Christopher D},
booktitle={Proceedings of the International Conference on Learning Representations (ICLR)},
year={2018}
}
@inproceedings{johnson2017clevr,
title={Clevr: A diagnostic dataset for compositional language and elementary visual reasoning},
author={Johnson, Justin and Hariharan, Bharath and Van Der Maaten, Laurens and Fei-Fei, Li and Lawrence Zitnick, C and Girshick, Ross},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={2901--2910},
year={2017}
}