A novel interpretable framework for cross-lingual content flagging, which significantly outperforms prior work both in terms of predictive performance and average inference time and can easily adapt to new instances without the need to retrain it from scratch.