Advertisement

'WebCrow 2.0' AI can solve crosswords in two languages

It will be pitted against human competitors in July.

Jenny Dettrick via Getty Images

Crossword puzzles aren't always easy to solve even for the most avid human fans, and they also remain one of the most challenging areas in artificial intelligence. Now, the University of Siena in Italy and expert.ai have a launched an AI software called WebCrow 2.0 that can solve crossword puzzles not just in English, but also in Italian. WebCrow 2.0 uses natural language processing technology to understand a puzzle's clues like a human player would.

That's trickier than it sounds, seeing as the same word could mean totally different things based on context, and crossword puzzle clues could contain a play on words. The answer for the clue "liquid that does not stick," for instance, is "scotch," which alludes to Scotch tape. Expert.ai's knowledge graph also gives it the reasoning power to find the correct meaning of words. Plus, the AI derives information from previously solved puzzles and its self-updating web knowledge to find the correct answer.

Last year, an AI called Dr. Fill outscored most of the best human competitors at the American Crossword Puzzle Tournament. While it's performance wasn't flawless — it did make three mistakes and was thwarted by a phonetically themed puzzle in one instance — Dr. Fill could solve puzzles faster than any human competitor. From July 18th through the 23rd, the creators of WebCrow 2.0 are also pitting their AI against human players. It will be a multilingual competition featuring previously unpublished crosswords in English and Italian and will demonstrate how good the AI actually is.

Marco Gori, a professor for University of Siena's Department of Information Engineering and Mathematical Sciences said:

"Can machines solve these as well as humans? How do they compare definitions and answer clues with niche or abstract references? Can they pick up on plays on words, linguistic nuances and even humor? We're ready to demonstrate how leveraging context can enable humans and software to work together and take AI-based cognitive abilities to new levels"