Відправити по sms: On guaranteed optimal robust explanations for NLP models