There may be some difficulty in doing this reliably. The problem is that PDF is a presentation format which attaches importance to good typography. Suppose you just wanted to output a single word: Tap.
A PDF rendering engine might output this as 2 separate calls, as shown in this pseudo-code:
moveto (x1, y); output ("T")
moveto (x2, y); output ("ap")
This would be done because the default kerning (inter-letter spacing) between the letters T and a might not be acceptable to the rendering engine, or it might be adding or removing some micro space between characters to get a fully justified line. What this finally results in is that the actual text fragments found in PDF are very often not full words, but pieces of them.