Yeah that's an interesting question. How to parse the content into meaningful pieces and then hash in such a way that the content is not known, but the hash can be mapped to where it was in the document at an earlier time.
Keep in mind that at the scale of a large work, some level of looseness may be sufficient. Identity and integrity are distinct, the former is largely based on metadata (author, title, publication date, assigned identifiers such as ISBN, OCLC, DOI, etc.). Integrity is largely presumed unless challenged.
