Discourse-Aware Graph Networks (DAGNs)

News

• Discourse-Aware Graph Networks for Textual Logical Reasoning is accepted to IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)!
• DAGN: Discourse-Aware Graph Network for Logical Reasoning is accepted to NAACL 2021!

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 the Extended DAGNs

sh run_extended_dagns.sh

Run DAGN

sh run_dagn.sh
sh run_dagn_aug.sh

sh logiqa_run_dagn.sh
sh logiqa_run_dagn_aug.sh

Run pre-trained LM baseline

sh run_roberta_large.sh
sh logiqa_run_roberta_large.sh

Citations

If you find any part of our papers or code is helpful, please generously cite with:

@ARTICLE{10136812,
    author={Huang, Yinya and Liu, Lemao and Xu, Kun and Fang, Meng and Lin, Liang and Liang, Xiaodan},
    journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
    title={Discourse-Aware Graph Networks for Textual Logical Reasoning}, 
    year={2023},
    volume={45},
    number={10},
    pages={11668-11688},
    keywords={Cognition;Task analysis;Representation learning;Transformers;Semantics;Linguistics;Birds;Natural language processing;logical reasoning;question answering;multi-turn dialogue reasoning;graph neural networks;supervised learning;zero-shot learning},
    doi={10.1109/TPAMI.2023.3280178}}

@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}
}