Reinforcement Learning from Human Feedback (RLHF) and derivative techniques like
Direct Preference Optimization (DPO) are task-alignment algorithms used to repurpose …

Large language models (LLMs) have shown great potential in natural language processing
tasks, but their application to machine translation (MT) remains challenging due to …

Automatic metrics for evaluating translation quality are typically validated by measuring how
well they correlate with human assessments. However, correlation methods tend to capture …