The Berkeley Crossword Solver – The Berkeley Synthetic Intelligence Analysis Weblog


We just lately constructed the Berkeley Crossword Solver (BCS), the primary pc program to beat each human competitor on the planet’s prime crossword match. The BCS combines neural query answering and probabilistic inference to attain near-perfect efficiency on most American-style crossword puzzles, just like the one proven beneath:

Determine 1: Instance American-style crossword puzzle

Crosswords are difficult for people and computer systems alike. Many clues are obscure or underspecified and might’t be answered till crossing constraints are taken under consideration. Whereas some clues are just like factoid query answering, others require relational reasoning or understanding tough wordplay.

Listed below are a handful of instance clues from our dataset (solutions on the backside of this publish):

  • They’re given out at Berkeley’s HAAS College (4)
  • Winter hrs. in Berkeley (3)
  • Area ender that UC Berkeley was one of many first faculties to undertake (3)
  • Angeleno at Berkeley, say (8)

The BCS makes use of a two-step course of to resolve crossword puzzles. First, it generates a likelihood distribution over attainable solutions to every clue utilizing a query answering (QA) mannequin; second, it makes use of probabilistic inference, mixed with native search and a generative language mannequin, to deal with conflicts between proposed intersecting solutions.

Determine 2: Structure diagram of the Berkeley Crossword Solver

The BCS’s query answering mannequin relies on DPR [Karpukhin et al., 2020], which is a bi-encoder mannequin sometimes used to retrieve passages which might be related to a given query. Somewhat than passages, nonetheless, our method maps each questions and solutions right into a shared embedding area and finds solutions immediately. In comparison with the earlier state-of-the-art technique for answering crossword clues, this method obtained a 13.4% absolute enchancment in top-1000 QA accuracy. We performed a handbook error evaluation and located that our QA mannequin sometimes carried out effectively on questions involving data, commonsense reasoning, and definitions, but it surely typically struggled to know wordplay or theme-related clues.

After operating the QA mannequin on every clue, the BCS runs crazy perception propagation to iteratively replace the reply chances within the grid. This permits data from excessive confidence predictions to propagate to more difficult clues. After perception propagation converges, the BCS obtains an preliminary puzzle answer by greedily taking the very best chance reply at every place.

The BCS then refines this answer utilizing a neighborhood search that tries to interchange low confidence characters within the grid. Native search works by utilizing a guided proposal distribution during which characters that had decrease marginal chances throughout perception propagation are iteratively changed till a domestically optimum answer is discovered. We rating these alternate characters utilizing a character-level language mannequin (ByT5, Xue et al., 2022), that handles novel solutions higher than our closed-book QA mannequin.

Determine 3: Instance adjustments made by our native search process

We evaluated the BCS on puzzles from 5 main crossword publishers, together with The New York Occasions. Our system obtains 99.7% letter accuracy on common, which jumps to 99.9% when you ignore puzzles that contain uncommon themes. It solves 81.7% of puzzles with out a single mistake, which is a 24.8% enchancment over the earlier state-of-the-art system.

Determine 4: Outcomes in comparison with earlier state-of-the-art Dr. Fill

The American Crossword Puzzle Event (ACPT) is the most important and longest-running crossword match and is organized by Will Shortz, the New York Occasions crossword editor. Two prior approaches to pc crossword fixing gained mainstream consideration and competed within the ACPT: Proverb and Dr. Fill. Proverb is a 1998 system that ranked 213th out of 252 opponents within the match. Dr. Fill’s first competitors was in ACPT 2012, and it ranked 141st out of 650 opponents. We teamed up with Dr. Fill’s creator Matt Ginsberg and mixed an early model of our QA system with Dr. Fill’s search process to win first place within the 2021 ACPT in opposition to over a thousand opponents. Our submission solved all seven puzzles in below a minute, lacking simply three letters throughout two puzzles.

Determine 5: Outcomes from the 2021 American Crossword Puzzle Event (ACPT)

We’re actually excited concerning the challenges that stay in crosswords, together with dealing with tough themes and extra complicated wordplay. To encourage future work, we’re releasing a dataset of 6.4M query reply clues, a demo of the Berkeley Crossword Solver, and our code at

Solutions to clues: MBAS, PST, EDU, INSTATER


Please enter your comment!
Please enter your name here