While moral reasoning has emerged as a promising research direction for large language models (LLMs), achieving robust generalization remains a critical challenge. This challenge arises from the gap between what is said and what is morally implied. In this paper, we build on metapragmatic links and the moral foundations theory to close the gap. Specifically, we develop a pragmatic-inference approach that facilitates LLMs, for a given moral situation, to acquire the metapragmantic links between moral reasoning objectives and the social variables that affect them. This approach is adapted to three different moral reasoning tasks to demonstrate its adaptability and generalizability. Experimental results demonstrate that our approach significantly enhances LLMs' generalization in moral reasoning, paving the road for future research to utilize pragmatic inference in various moral reasoning tasks.

翻译：尽管道德推理已成为大型语言模型（LLMs）的一个重要研究方向，但实现稳健的泛化仍是一个关键挑战。这一挑战源于“所言”与“道德所寓”之间的差距。本文基于元语用链接与道德基础理论来弥合这一差距。具体而言，我们提出了一种语用推断方法，使LLMs能够在给定道德情境下，习得道德推理目标与影响这些目标的社会变量之间的元语用链接。该方法被适配至三种不同的道德推理任务，以展示其适应性与泛化能力。实验结果表明，我们的方法显著提升了LLMs在道德推理中的泛化性能，为未来研究在各种道德推理任务中运用语用推断铺平了道路。