What about a bubble-fill, with multiple bubbles per candidate? Instructions, beyond the standard how-to-darken-a-bubble stuff, amount to "fill in more bubbles for the candidate(s) you really like, fewer bubbles for the candidate(s) you don't like so much, and no bubbles for the candidate(s) you don't like at all." Maybe some examples. Vote-reader filters out the noise of exactly which bubbles were filled, then normalizes each ballot into a preference ranking, plus some implied trivia about preference strength.