# Random Text Perturbations Work, but not Always

arXiv: https://arxiv.org/abs/2209.00797 · HTML: https://mlsanity.com/paper/2209.00797/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 10 of its 18 references that are in the map.
- First posted on arXiv 2 Sep 2022.

## Builds on (most cited first, 10 of 10)

- [Character-level Convolutional Networks for Text Classification](https://mlsanity.com/paper/1509.01626/) (2015)
- [SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition](https://mlsanity.com/paper/1904.08779/) (2019)
- [Unsupervised Data Augmentation for Consistency Training](https://mlsanity.com/paper/1904.12848/) (2019)
- [EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks](https://mlsanity.com/paper/1901.11196/) (2019)
- [A Survey of Data Augmentation Approaches for NLP](https://mlsanity.com/paper/2105.03075/) (2021)
- [Contextual Augmentation: Data Augmentation by Words with Paradigmatic Relations](https://mlsanity.com/paper/1805.06201/) (2018)
- [An Empirical Survey of Data Augmentation for Time Series Classification with Neural Networks](https://mlsanity.com/paper/2007.15951/) (2020)
- [SwitchOut: an Efficient Data Augmentation Algorithm for Neural Machine Translation](https://mlsanity.com/paper/1808.07512/) (2018)
- [Sequence-to-Sequence Data Augmentation for Dialogue Language Understanding](https://mlsanity.com/paper/1807.01554/) (2018)
- [Text Augmentation in a Multi-Task View](https://mlsanity.com/paper/2101.05469/) (2021)

## Built on by (most cited first, 1 of 1)

- [Probabilistic Linguistic Knowledge and Token-level Text Augmentation](https://mlsanity.com/paper/2306.16644/) (2023)

## Abstract (arXiv)

We present three large-scale experiments on binary text matching classification task both in Chinese and English to evaluate the effectiveness and generalizability of random text perturbations as a data augmentation approach for NLP. It is found that the augmentation can bring both negative and positive effects to the test set performance of three neural classification models, depending on whether the models train on enough original training examples. This remains true no matter whether five random text editing operations, used to augment text, are applied together or separately. Our study demonstrates with strong implication that the effectiveness of random text perturbations is task specific and not generally positive.