I see most of my sites use anchor text that is the same as the title of the document, but the html/pdf titles are either garbage or generated and therefore the same every time. I looked over the "data from field" and noticed there is no option to do this (although I got excited at first seeing the URL anchor option, but it's for something else). I suspect that info is not sent over by the server so no way of holding on to it at crawl/fetch time. Any ideas other than writing a custom regex for finding the titles on each page (they are all different for each site and if the page designs are ever changed, all the regex has to be rewritten).
The other issue with link text is that you could have multiple links with different text, and knowing which one to use could be problematic.
There probably isn't a good way right now if the <title> tag isn't used properly on the site, and the pages aren't consistent.
A future update will have a better title guesser for PDFs by looking at the content of the page.