Multilingual machine translation models exhibit inherent task-specific modularity, which can be leveraged to mitigate interference and enhance knowledge transfer across languages.
