The table showing percentage of ? vs. ?$ seems to answer a question I didn't have (namely, of queries that have question marks, how many have one at the end?). I think I am more interested in knowing "Of queries that end with a question mark, how many of those had other question marks as well?" To me, that would distinguish a likely question from a likely other use of question marks. Would it be possible to compute those percentages?
Topic on User talk:TJones (WMF)/Notes/Dropping Final Question Marks in the Top 10 Wikipedias
Awesome and interesting question, Kevin! There aren't a ton of examples, even in 1M queries, but it looks like most are still questions or junk, with only a handful of possible wildcard patterns (all of which won't work anyway) :(
See the addendum, "Queries With Multiple Question Marks" (I can't get a link to the right section to work).
Thanks for the follow-up. So 90-99% of queries that end with a question mark have no other question marks in the query.
Your "turnip" example definitely looks like an attempt to solve a crossword puzzle.