Mitigating Bias in Automated Essay Scoring for ESL Learners via Contrastive Learning
October 6, 2026
Contrastive Learning with Matched Essay Pairs (CL-MEP) reduces scoring bias against high-proficiency ESL learners by 39.9% in automated essay scoring systems. The bi-directional alignment strategy disentangles syntactic complexity from surface-level grammatical errors to improve scoring accuracy while mitigating disproportionate penalties.