Utilising Explanations to Mitigate Robot Conversational Failures