pip install torch==1.3.1
In the fairseq/connotations folder you can put the train.source , train.target , val.source , val.target files The data and model is available on request as we understand that our data can be misused for generations and manipulation
Please email tuhin.chakr@cs.columbia.edu / smara@cs.columbia.edu and fill the form https://docs.google.com/forms/d/e/1FAIpQLSdM3qRWHV4lgpIkRkZ74YwS2haD6qUTrzujsasJpaBbH9EnQA/viewform?usp=sf_link
Change the encoder.json path to correct path in fairseq/fairseq/data/encoder/gpt2_bpe_utils.py line 131
run sh preprocess1.sh
run sh preprocess2.sh
run sh trainbart.sh
pip install allennlp==1.0.0
pip install allennlp_models
After model is finetuned use python finalent.py to generate
Test data and model output for ENTRUST and various systems is in fairseq/testdata/ folder
Input partisanfinal.source fallaciesfinal.source
Output for ENTRUST partisanfinal.hypo fallaciesfinal.hypo
Gold is fallaciesgoldhuman and partisangoldhuman
If you use any data or code please cite us
@article{chakrabarty2021entrust,
title={ENTRUST: Argument Reframing with Language Models and Entailment},
author={Chakrabarty, Tuhin and Hidey, Christopher and Muresan, Smaranda},
journal={arXiv preprint arXiv:2103.06758},
year={2021}
}