For formats that don't contain any embedded chapter information (such as plain text, Rich Text Format, and some variants of HTML and PDF), Stanza tries to automatically detect where chapters are. It usually does a good job, but its rules are not perfect, and it can sometimes make bizarre judgements about where a chapter break is. Aside from looking strange and hindering chapter-based navigation, there are no other adverse effects of mis-identified chapters. If you are willing to share the offending text with us by mailing it to support@lexcycle.com, we are always looking for data that will help us improve our chapter identification heuristics.