Date: March 18, 2012
Byline: Steve Lohr
In Crosswords, Itís Man Over Machine, for Now
Score one for humans and their subtle, quirky, pattern-matching brains.
Over the weekend, an impressive crossword-solving computer program, called Dr. Fill, which I wrote about earlier, matched its digital wits against the wetware of 600 of the nation’s best human solvers at the American Crossword Puzzle Tournament in Brooklyn.
Before the tournament, Matthew Ginsberg, the creator of Dr. Fill and an expert in artificial intelligence, predicted a range of likely outcomes for his clever code. In simulations of 15 past tournaments, Dr. Fill finished on top three times. But at other times it stumbled.
Dr. Ginsberg said the program would probably finish in the top 50, among the 600 contestants. With only seven puzzles, the range of possible outcomes was wide. “If I’m lucky, I’ll win,” he said, “If I’m unlucky, I’ll end up 150th,” but still in the top fourth.
In the weekend tournament, Dr. Fill finished 141st, or would have (only human solvers got official rankings). “It was within the range, but I wish it had done better,” Dr. Ginsberg said on Sunday. “I’ll be back next year.”
Dr. Fill typically thrives on conventional crosswords, even ones with arcane clues and answers. Indeed, the seventh puzzle, a difficult one, it got perfectly.
But the computer program is literal minded, and tends to struggle on puzzles with humor, and puzzles with unusual themes or letter arrangements.
“Two of the puzzles were bizarre in ways that were bad for it,” Dr. Ginsberg explained.
One included several words that had to be spelled backwards. And another puzzle required correct answers, not just across and down, but some words arrayed diagonally as well. Those were the second and fifth puzzles in the tournament.
“Dr. Fill got killed on puzzles two and five,” said Will Shortz, director of the tournament, who is also the crossword puzzle editor of The New York Times.
Mr. Shortz commissions and edits the puzzles in the contest. Dan Feyer, last year’s winner, said before the tournament began that he figured Mr. Shortz would include a puzzle or two that involved innovative twists or patterns to trip up Dr. Fill.
Did Mr. Shortz choose puzzles with foiling the computer program in mind? Mr. Shortz shook his head no, and smiled, when asked that question after the tournament ended Sunday afternoon.
“I have no reason to believe that’s the case,” Dr. Ginsberg said. “But people did say the puzzles were particularly innovative this year. If so, that’s good for the crossword puzzle community” — though a greater programming challenge for Dr. Fill’s creator.
Dr. Ginsberg himself constructs crossword puzzles as a hobby, with more than two dozen published in The New York Times.
In a further sign of crossword continuity, this year’s winner was Mr. Feyer, once again, for the third straight time.