We describe a hierarchical approach to modeling text that allows machine-actionable canonical citation of text at many levels of specificity. This model address the problem of overlapping or mutually exclusive analyses. In turn, this flexibility in citation allows rich linking of textual transcriptions and other data to regions-of-interest on digital images, of particular value to codicological and paleographic study. Our examples are from work on Byzantine manuscripts containing Greek epic poetry and scholarly commentary, but our approach can apply to any image-based project in documenting books, manuscripts, inscriptions or other text-bearing surfaces.