How do we add support for keeping book-nlp loaded (and Stanford NLP which it uses), so we don't have to reload every time we process a new book?