Exploring the pathways of our lexicon
(Mostly Human) Crossword Whizzes Head to Brooklyn
This weekend, puzzlers will come together in Brooklyn for the 35th American Crossword Puzzle Tournament, organized by New York Times puzzle editor Will Shortz. The reigning champ, Dan Feyer, has been described as a crossword-solving machine. But he better look out, because this time there will be competition from an actual crossword-solving machine.
Feyer has won the last two ACPT championships in impressive fashion (taking over from the equally impressive Tyler Hinman, who had won the previous five tournaments). But along with these illustrious humans, watch out for a silicon-based competitor named Dr. Fill, a crossword-solving computer program developed by Matt Ginsberg.
As I describe in my column in last Sunday's Boston Globe, Dr. Fill isn't formally competing in the ACPT, but the humans in the room will be keenly aware of the artificial intelligence solving puzzles at the same time as them at the Brooklyn Bridge Marriott. And anyone whose score is better than Ginsberg's software will win a button that says "I Beat Dr. Fill." Will Shortz told me he's ordered 150 buttons, just to be safe, but I don't think he'll be giving out too many... if any. Ginsberg has been making steady improvements to Dr. Fill's algorithms, and the program could easily outpace the human competitors unless it gets stumped by some particularly tricky wordplay in one or more puzzles.
We've been covering the ACPT for a few years now, beginning with Tyler Hinman's fifth consecutive victory in 2009. Visual Thesaurus puzzlemaster Brendan Emmett Quigley has given us the rundown on the remarkable rise of "Steely" Dan Feyer, who ran away with it in 2010 and again in 2011. Brendan has once again agreed to provide commentary on all the action this year, so look for his reports on how things progress on Saturday and Sunday in this space.
The presence of Dr. Fill adds an interesting wrinkle to the proceedings this year, but the carbon-based competition should be plenty dramatic. Feyer and Hinman are favorites again, of course, but there should be some up-and-coming solvers that will give them a run for their money. Sadly, three-time winner Trip Payne has decided to retire from the competition, so he won't be gracing the finals as he has so many years in the past. (Those who have seen the documentary Wordplay, which showcased the exciting 2005 tournament, may recall that Payne is the last ACPT winner not named Hinman or Feyer.) Anne Erdmann, who placed third in the last two Feyer-dominated contests, will be looking to move even higher. Perennial threats like Francis Heaney and Howard Barkin can't be counted out. And we could see the emergence of a new wunderkind, David Plotkin, who won the B division last year at the tender age of 21 and will be up against the top solvers in the A division this year.
As in past years, contestants will work through seven rounds of puzzles: three on Saturday morning, three on Saturday afternoon, and one on Sunday morning. The crosswords vary in size and difficulty, with Puzzle #5 traditionally being the hardest. Scores are calculated according to speed and accuracy. A solver gets 10 points for every correct word, a 25-point bonus for each full minute beating the suggested solution time, and a 150-point bonus for an error-free solution. Each wrong letter reduces the score by 25 points. The three contestants with the highest combined score on the seven puzzles get to advance to Sunday's final round, solving a tough crossword on stage wearing noise-blocking headphones.
If Dr. Fill does end up beating the humans, it will have to make up in speed for any errors that creep in because of its lack of human understanding of creative wordplay. (Puzzle #5, which usually involves a theme that even smart humans struggle with, may make life very difficult for Dr. Fill.) So if the program makes an error in each of two puzzles, then it will have to compensate for missing out on 300 bonus points by solving all the puzzles faster than the fastest human by an average of at least two or three minutes. I can attest that this is completely in the realm of possibility, because Ginsberg has let me play around with the Dr. Fill program. (Amazingly, it can run on a MacBook Pro notebook computer, quite unlike the massive Watson supercomputer that IBM used to beat the humans on "Jeopardy!")
If you have qualms about Dr. Fill possibly outdoing the mere mortals in the room, listen to these words of equanimity from Will Shortz about Ginsberg's creation: "I have no compunctions about Dr. Fill's crossword-solving prowess," he told me. "It's not as if Dr. Fill can 'think.' It can only do what Matt has programmed it to do. I feel pride in the human ingenuity that went into Dr. Fill's programming." So either way, humans win!