The page segmentation file (.pseg.png) contains pixel accurate information about where the lines are.
For finding words within lines, it depends on the recognizer and the script of how to get at that information. ocropus-lattices gives you bounding boxes relative to the text line. ocropus-rpred (the new recognizer) outputs a sequence of classification vectors that you could use.
Tom