DAGN: Discourse-Aware Graph Network for Logical Reasoning (NAACL'21)

Official implementation for the NAACL'21 short paper DAGN: Discourse-Aware Graph Network for Logical Reasoning.

Dependencies

This code has been tested with the following dependencies and versions:

python==3.7.9
torch==1.5.0
transformers==3.1.0
numpy==1.19.2
gensim==3.8.3

Preparation

The ReClor data is ready in the ./reclor_data. To run the LogiQA data, create ./logiqa_data where you put the downloaded data.

Run pre-trained LM baseline

sh run_roberta_large.sh
sh logiqa_run_roberta_large.sh

Run DAGN

sh run_dagn.sh
sh run_dagn_aug.sh

sh logiqa_run_dagn.sh
sh logiqa_run_dagn_aug.sh

Citation

If any part of our paper or code is helpful, please generously cite with:

@InProceedings{zhang2021video,
author = {Huang, Yinya and Fang, Meng and Cao, Yu and Wang, Liwei and Liang, Xiaodan},
title = {DAGN: Discourse-Aware Graph Network for Logical Reasoning},
booktitle = {NAACL},
year = {2021}
}