Are Embedding Spaces Interpretable? Results of an Intrusion Detection Evaluation on a Large French Corpus