Documents with a font encoding of 'Identity-H' would seem to warrant special attention. This encoding usually indicates a particular kind of mix-and-match of characters, resulting in problems such as regular hyphens being rendered with the Unicode U+00AD 'soft hyphen'.

This affects 54 files in our current corpus. It would seem to be related to the "CID" encoding mentioned below; see also http://indesignsecrets.com/cid-identity-h-fonts-are-back.php

On many stretches of text where pdfBox is unable to map glyphs to normative code points, it instead numbers the glyphs and outputs them in the form "g81g72g81g3g82g80g3g76g81g86g87" etc. The letter may be g, a, x etc. It may also come out as strings like 'glyph', 'afii' (in connection with Arabic glyphs) or 'cid' (seen only in documents made on a Mac).

Most of these unmapped glyphs can be found with the regex  \d(([A-Za-z]+)\d+){3}  — this matches runs of three of characters and digits right next to each other. A high proportion of text on this form indicates that the document is mostly unreadable (or entirely concerned with chemical reactions; some false positives like this are to be expected).

Classification of documents, sorted by score: http://heim.ifi.uio.no/olasba/nora/anonglyphs06.log

(The middle column is the string or letter that most frequently marks out the unmapped glyphs.)

Source code: http://heim.ifi.uio.no/olasba/nora/findanonglyph06.pl.txt

• ^{—Ola, 13 Oct}