Aligning Multilingual Reasoning with Verifiable Semantics from a High-Resource Expert Model

Open in new window