MET: Theory-Grounded and Culture-Aware Multilingual Moral Reasoning

Researchers introduce MET (Multilingual Ethics with Theory-grounded reasoning), a method for improving moral decision-making in diverse linguistic contexts by addressing gaps in multilingual evaluation, inference scaffolds, and training supervision.

The work presents three key contributions: MCLASH, a benchmark capturing culturally situated moral intuitions; MET, a two-step prompting method using expert-curated grounds from psychology and philosophy; and MET-D, a self-distillation training stage that requires no external supervision.

MET-D improves macro-F1 over base models (Qwen3-4B, Qwen3-8B, Gemma3-4B) by an average of 3.71 points on MCLASH and 4.23 on MMoralExceptQA, while increasing native-language reasoning by 62.13 points on average.

These contributions open the path for culture-aligned, theory-grounded multilingual moral reasoning.