The paper introduces MCLASH, a multilingual benchmark designed around culturally situated moral intuitions and social norms, plus MET, a two-step prompting method grounded in expert-curated psychological and philosophical theories. MET-D adds self-distillation without external supervision. Across Qwen3-4B, Qwen3-8B, and Gemma3-4B, MET-D improves macro-F1 by an average of 3.71 points on MCLASH and 4.23 points on MMoralExceptQA. The largest reported MCLASH gain is 12.94 points for Malay on Qwen3-8B, while native-language reasoning rises by 62.13 points on average.
No heat snapshots are available in the last 24 hours.