Abstract Despite the importance of propositional logic in arti?cial intelligence, the notion of language independence in the propositional setting (not to be confound with syntax independence) has not received much attention so far. In this paper, we de?ne language independence for a propositional operator as robustness w.r.t. symbol translation. We provide a number of characterizations results for such translations. We motivate the need to focus on symbol translations of restricted types, and identify several families of interest. We identify the computational complexity of recognizing symbol translations from those families. Finally, as a case study, we investigate the robustness of belief revision/merging operators w.r.t. translations of different types. It turns out that rational belief revision/merging operators are not guaranteed to offer the most basic (yet non-trivial) form of language independence; operators based on the Hamming distance do not suffer from this drawback but are less robust than operators based on the drastic distance.