It's Morphin' Time! Combating Linguistic Discrimination with Inflectional Perturbations
This addresses bias in NLP for minority language speakers, but it is incremental as it builds on existing adversarial methods.
The paper tackles linguistic discrimination in NLP models by exposing biases against non-standard English varieties through inflectional perturbations, showing that adversarially fine-tuning models like BERT for one epoch improves robustness without harming clean data performance.
Training on only perfect Standard English corpora predisposes pre-trained neural networks to discriminate against minorities from non-standard linguistic backgrounds (e.g., African American Vernacular English, Colloquial Singapore English, etc.). We perturb the inflectional morphology of words to craft plausible and semantically similar adversarial examples that expose these biases in popular NLP models, e.g., BERT and Transformer, and show that adversarially fine-tuning them for a single epoch significantly improves robustness without sacrificing performance on clean data.