Request Correction

C4: A Multilingual Benchmark for Retrieval-Augmented Generation Based on the Catechism of the Catholic Church and Its Compendium

We introduce a new multilingual case study for evaluating retrieval augmented generation (RAG) systems, based on the Catechism of the Catholic Church and its Compendium. The Catechism is a structured document with numbered paragraphs, officially translated into many languages under strict editorial alignment. The Compendium reformulates this material into a question-answer format with explicit citations to the corresponding paragraphs. Together, they form a set of parallel monolingual corpora that share identical semantic structure, enabling direct, controlled comparison of RAG performance across languages. Beyond its theological origin, this text pair closely mirrors real-world applications of RAG in institutional contexts, such as querying internal policy documents with associated FAQ-style summaries, making it a practical testbed for multilingual retrieval and grounded answer generation. We release our data collection scripts and baseline results for further research.