Enviar por SMS: On guaranteed optimal robust explanations for NLP models