Low-resource languages are an important and challenging application for DA, typically for neural machine translation (NMT). Using WordNet, backtranslation and etc is proved to be successful on Low-Resource NMT.

University of Michigan - Ann Arbor

It's widely used in Low-Resource NMT, Mitigating Bias, Fixing Class Imbalance, Few-shot Learning, and Adversarial Examples (AVEs).

Applications of DA in NLP

New instances can be obtained based on existing ones by modifying the features with transformations that do not change the label. To go beyond the token level and add more diversity to the augmented sentences, data augmentation can also be performed on sentence parts.

Data Augmentation for Low-Resource NLP

https://arxiv-download.xixiaoyao.cn/pdf/2105.03075.pdf

A Survey of Data Augmentation Approaches for NLP

DA in Low-Resource NMT

Various bias can be learned by LM when it's trained on the data containing human discrimination. Researchers have tried using fixed rules or specific models to eliminate bias and used it for mitigating bias contained in the LM.

DA in Mitigating Bias

Fixing class imbalance typically involves a combination of undersampling and oversampling.

DA in Fixing Class Imbalance

DA methods can ease few-shot learning by adding more examples for novel classes introduced in the few-shot phase. 

DA in Few-shot Learning

Adversarial examples can be generated using innocuous label-preserving transformations (e.g. paraphrasing) that fool state-of-the-art NLP models

DA in AVEs

1. Decrease the requirement of in-depth linguistic insight.
2. How to achieve cross-tasks data augmentation?

Open Issues of Applying Data Augmentation to Low-resource NLP

1. Distant & weak supervision.
2. Cross-Lingual Annotation Projections
3. Learning with Noisy Labels
4. Non-Expert Support

Learn Before

Related