Reproduce “A Survey of Data Augmentation Approaches for NLP” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Steven Y. Feng, Varun Gangal, Jason Wei, Sarath Chandar, Soroush Vosoughi, Teruko Mitamura, Eduard Hovy

This comprehensive survey from Carnegie Mellon University, Google Research, and Mila categorizes and analyzes data augmentation techniques for Natural Language Processing, addressing the challenge of data scarcity in NLP. It establishes a structured taxonomy for methods like rule-based, example interpolation-based, and model-based approaches, detailing their applications across tasks from low-resource language support to bias mitigation.

Key insights

  • The discrete nature of language presents a fundamental challenge for data augmentation, making it difficult to directly apply techniques successful in computer vision while maintaining semantic and syntactic coherence.
  • There is a clear trade-off between the implementation simplicity of data augmentation methods (e.g., rule-based) and the magnitude of performance improvements they offer (often greater with model-based approaches).
  • A significant gap exists in the theoretical understanding of why data augmentation is effective in NLP, hindering the development of more principled and universally robust techniques.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.