Here's an example where highlighting changed words in a comment is kinda distracting. I think the right solution is to compute a levenshtein distance and only highlight changed words if the lines are sufficiently similar.