I agree with derby. Your data structure should explicitly model the whole book, including the text, the chapter numbers, and the paragraph numbers. It's better not to infer chapter numbers and paragraph numbers implicitly from their positions within a Perl nested data structure. A well-designed data structure will be extensible. If you decide later you want to model another attribute of the book—say, sentence numbers—you can do so without having to re-implement the data structure or create a new one. The data structure will then also map neatly to an external, reusable representation such as XML.