[ad_1]
Last week American Crossword Tournament, A spectacular competitor held as a virtual event with more than 1,000 participants, gave the news. (And, although I finished 143rd, unfortunately I wasn’t me.) For the first time, artificial intelligence managed to outperform human solutions in the race to fill networks with speed and accuracy. It was a victory for Dr. Fill, an automaton for solving crossword puzzles, who has been competing against carbon-based crossbowmen for nearly a decade.
For some observers, it may seem like another area of human endeavor that currently dominates AI. Report on Dr. Fill’s achievement For slate, Oliver Roeder wrote: “Checkers, backgammon, chess, Go, poker and other games have witnessed machine invasions in the main AI that have fallen one by one. Now crossword puzzles have joined them.” But the study of how Dr. Fill accomplished this feat shows much more than the recent struggle between humans and computers.
IBM’s Watson supercomputer surpassed Ken Jennings and Brad Rutter Danger! a little more than 10 years ago, Jennings he replied, “Me, we welcome our mainframe computers.” But Jennings was too early to throw in the towel for humanity. As of now, the latest advances in IA show not only the computational understanding of natural language, but also its limitations. In the case of Dr. Fill, his performance informs us of the intelligence that humans bring to the unique linguistic challenge of solving a crossword puzzle, in line with the inventive souls who invent puzzles. In fact, if we look at how the software crosses the crossword puzzle track, it reveals what our brains are doing when we play with language.
Fill was a doctor Matt Ginsberg, a computer scientist, is also a published crossword builder. Since 2012, is informally joining Dr. ACPT, making additional improvements to its resolution software each year. This year, however, Ginsberg joined forces Berkeley Natural Language Treatment Group, Consisting of graduate and postgraduate students led by UK Berkeley professor Dan Klein.
Klein and his students began serious work on the project in February, and later turned to Ginsberg to see if they could reconcile their efforts for this year’s tournament. Two weeks before the start of ACPT, the hybrid system was hacked together, with the Berkeley group interpreting traces of the neural network methods in conjunction with Ginsberg’s code to effectively comply with the crossword grid.
(For anyone interested in spoilers beforehand Solve ACPT puzzles after the event.)
The new and better Dr. Fill fills the grid with pil-pil activity (you can see it in action here). But the truth is that the program is very methodical, examines a clue and gives an initial ranking of the list of candidates to get the answer, and then reduces the chances based on factors like how they fit in with other answers. The correct answer may be deep in the list of candidates, but it allows enough context to fit all the way to the top.
Dr. Fill is trained in data collected from past crossword puzzles that have appeared in various outlets. To solve a puzzle, the program cites clues and answers that it has already “seen”. Like humans, Dr. Fill must build on what he has learned in the past in the face of a new challenge, seeking connections between new and old experiences. For example, the second puzzle of the competition Wall Street Journal Mike Shenk, the editor of the crossword puzzle, focused on a topic to complete new fantasy sentences that added the letters -ITY in long answers, such as turning OPIUM DENS into OPIUM DENSITY (expressed as “Poppy product of the potency of a power?”). Dr. Fill was fortunate that, despite the unusual phrases, some of the answers appeared in a crossword puzzle similar to the one published in 2010. in The Los Angeles times, Which Ginsberg included in its database of more than 8 million tracks and responses. As the crossword puzzles in the tournament were quite different, Dr. Fill was challenged to give the correct answers. (OPIUM DENSITY, for example, was referred to in 2010 as the “Neighborhood Drug Trafficking Measure?”)
For all answers, whether or not they are part of the puzzle theme, the program offers thousands of options for creating candidates that would best fit the tracks, sorting them by probability and checking them according to network boundaries, such as interacting with entries below. Sometimes the main candidate is appropriate: for the “Imposing Group” clue, for example, Dr. Fill preferred the correct answer, ARRAYS, as the word. The word “imposer” never appeared in the previous traces of the word, but so did other synonymous words like “spectacular,” allowing Dr. Fill to infer a semantic link.
[ad_2]
Source link
