Mitigating Bias in Automated Essay Scoring for ESL Learners via Contrastive Learning
The study used contrastive learning to reduce bias in automated essay scoring (AES) systems for English as a Second Language (ESL) learners. By training a model with 'Fairness Triplets,' the researchers significantly decreased the score disparity between native and ESL students by nearly 40%, showing that contrastive learning effectively mitigates unfair linguistic penalties.