Simple and effective data augmentation for compositional generalization

Yuekun Yao, Alexander Koller


Abstract
Compositional generalization, the ability to predict complex meanings from training on simpler sentences, poses challenges for powerful pretrained seq2seq models. In this paper, we show that data augmentation methods that sample MRs and backtranslate them can be effective for compositional generalization, but only if we sample from the right distribution. Remarkably, sampling from a uniform distribution performs almost as well as sampling from the test distribution, and greatly outperforms earlier methods that sampled from the training distribution.We further conduct experiments to investigate the reason why this happens and where the benefit of such data augmentation methods come from.
Anthology ID:
2024.naacl-long.25
Volume:
Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)
Month:
June
Year:
2024
Address:
Mexico City, Mexico
Editors:
Kevin Duh, Helena Gomez, Steven Bethard
Venue:
NAACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
434–449
Language:
URL:
https://aclanthology.org/2024.naacl-long.25
DOI:
Bibkey:
Cite (ACL):
Yuekun Yao and Alexander Koller. 2024. Simple and effective data augmentation for compositional generalization. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 434–449, Mexico City, Mexico. Association for Computational Linguistics.
Cite (Informal):
Simple and effective data augmentation for compositional generalization (Yao & Koller, NAACL 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.naacl-long.25.pdf
Copyright:
 2024.naacl-long.25.copyright.pdf