process-aligned

This paper proposes PAMT, a process-aligned reinforcement learning framework for multi-domain machine translation that combines domain-aware long chain-of-thought supervision with step-level process rewards to improve domain-sensitive translation decisions.