Does a Large Language Model Really Speak in Human-Like Language?