End-to-End Self-Debiasing Framework for Robust NLU Training
Computation and Language
2021-09-07 v1
Abstract
Existing Natural Language Understanding (NLU) models have been shown to incorporate dataset biases leading to strong performance on in-distribution (ID) test sets but poor performance on out-of-distribution (OOD) ones. We introduce a simple yet effective debiasing framework whereby the shallow representations of the main model are used to derive a bias model and both models are trained simultaneously. We demonstrate on three well studied NLU tasks that despite its simplicity, our method leads to competitive OOD results. It significantly outperforms other debiasing approaches on two tasks, while still delivering high in-distribution performance.
Cite
@article{arxiv.2109.02071,
title = {End-to-End Self-Debiasing Framework for Robust NLU Training},
author = {Abbas Ghaddar and Philippe Langlais and Mehdi Rezagholizadeh and Ahmad Rashid},
journal= {arXiv preprint arXiv:2109.02071},
year = {2021}
}
Comments
Findings ACL 2021