README.md

This repository contains code of Jung et al's ACL 2020 paper titled "Posterior Calibrated Training on Sentence Classification Tasks"

Requirements

We use python 3.7. Please run pip install -r requirement.txt to install python dependencies.

Running the BERT classifier with PosCal training

Note that you add the flag --poscal_train for PosCal training. The example below uses the ShortRomance dataset in xSLUE (Kang et al., 2019). Please refer to the xslue resposiotry to download the datasets.

python classify_bert.py \
    --model_type bert \
    --model_name_or_path bert-base-uncased \
    --task_name ShortRomance \
    --do_train \
    --poscal_train \
    --data_dir $PATH/to/the/data \
    --output_dir $PATH/to/the/output

Citation

@inproceedings{jung20acl_poscal,
    title = {Posterior Calibrated Training on Sentence Classification Tasks},
    author = {Taehee Jung, Dongyeop Kang, Hua Cheng, Lucas Mentch, and Thomas Schaaf},
    booktitle = {2020 Annual Conference of the Association for Computational Linguistics (ACL)},
    url = {https://arxiv.org/abs/2004.14500},
    year = {2020}
}