PDFs can be produced by mapping completely unrelated characters to glyphs that appear as actual characters (basically, they'd embed a "compacted" font that only

PDFs can be produced by mapping completely unrelated characters to glyphs that appear as actual characters (basically, they'd embed a "compacted" font that only has the glyphs required for the document, and then map them to ASCII or something). This was quite common with pdftex documents not using ASCII characters in the past, thus making text unsearchable (and even more so when going the ps2pdf route). For example, you'd have a Cyrillic document in which when you selected text in, highlighted text would be some jumbled ASCII characters. The fact is that PDF can display one thing and have underlying semantic text be something else entirely (frequently used for OCRing: you show actual scanned images of text, and put the invisible OCRed text as searchable ). It works in the other direction too: you could solve the hyphenation problem in the same way by having PDF include invisible non-hyphenated word in place of the hyphenated one for searching. Still, PDF is mostly a laying-out format, and while tools have evolved to provide some "meaning" to rendered content, it is never going to be semantic in the sense markup languages can be (i.e. there is no "emphasis", "quote" or "header" command for PDF, instead, it just uses a different font). To put things into perspective, TeX files can be semantic (if a semantic TeX .fmt like LaTeX is used) like HTML/ePub, but PDF is an output format, just like DVI is.