Multilingual watermarking aims to make large language model (LLM) outputs traceable across languages, yet current methods still fall short. Despite claims of cross-lingual robustness, they are evaluated only on high-resource languages. We show that existing multilingual watermarking methods are not truly multilingual: they fail to remain robust under translation attacks in medium- and low-resource languages. We trace this failure to semantic clustering, which fails when the tokenizer vocabulary contains too few full-word tokens for a given language. To address this, we introduce STEAM, a detection method that uses Bayesian optimisation to search among 133 candidate languages for the back-translation that best recovers the watermark strength. It is compatible with any watermarking method, robust across different tokenizers and languages, non-invasive, and easily extendable to new languages. With average gains of +0.23 AUC and +37% TPR@1%, STEAM provides a scalable approach toward fairer watermarking across the diversity of languages.

翻译：多语言水印旨在使大语言模型（LLM）的输出在不同语言中可追溯，但现有方法仍存在不足。尽管声称具有跨语言鲁棒性，但这些方法仅在高资源语言上进行了评估。我们证明，现有的多语言水印方法并非真正的多语种：在翻译攻击下，它们在中、低资源语言中无法保持鲁棒性。我们将此失败归因于语义聚类——当分词器词汇表中包含的完整单词标记数量过少时，语义聚类会失效。为解决这一问题，我们提出STEAM检测方法，利用贝叶斯优化在133种候选语言中搜索能最佳恢复水印强度的反向翻译。该方法与任意水印技术兼容，跨不同分词器和语言鲁棒，具有非侵入性，且易于扩展至新语言。平均AUC提升0.23、TPR@1%提升37%，STEAM为跨语言多样性的更公平水印提供了可扩展方案。