Inside the Box: Crossword Puzzle Constructing in the Computer Age

Crossword constructor Anna Shechtman, 24, sticks out like a worn eraser in a delete-button world. These days, most serious constructors use computer software and word databases to make their puzzles, at least to some degree—but Shechtman writes hers by hand.

“I have totally gone halfway through a puzzle, or three-quarters of the way through a puzzle, and found that it doesn’t work and basically had to start all over,” she says. “And yet, there’s a part of me that doesn’t want to cede that [effort] to the software.”

But Shechtman, whose puzzles have appeared in both mainstream and independent publications such as The New York Times and the American Values Club crossword, agrees that computers can be an asset. “The majority of people I know and respect in the field use software,” she says, and “they’re turning out such phenomenal puzzles.”

Indeed, over the past 15 years, the transition from pencil and paper to computer screen has improved efficiency and abetted creativity, according to devoted cruciverbalists. By combining human ingenuity with algorithmic aptitude, serious constructors are pushing the bounds of the box with livelier words and other conceits that the British journalist Arthur Wynne could never have foreseen when he published the first American crossword more than 100 years ago.

“The standards have risen,” says Will Shortz, editor of The New York Times’ puzzle. “We expect more because of computer assistance now.”

There’s a scene in the 2006 documentary Wordplay, about crossword puzzle aficionados, in which veteran constructor Merl Reagle demonstrates how to create a puzzle by hand with a pencil and graph paper, filling in entries, shading in black squares, and plotting which letters make sense where.

That scene intrigued constructor David Steinberg, who saw his first New York Times puzzle published at age 14. “He made it look so easy,” recalls Steinberg, now 17. “So I decided to go up the next day to my room, and I took out a piece of graph paper, and I tried to do what Merl did, and of course, it didn’t come out as well.” But he submitted his attempt to the Times anyway. “It was promptly rejected,” he says.

After about nine “Nos,” Steinberg decided to try a different method, so he turned to a software program called Crossword Compiler—one of the best known in the biz. (Crossfire is another. Crossword Compiler is PC-compatible; Crossfire is Mac-friendly.) The program helped him break into the game.

When writing a crossword, a constructor typically “seeds” the grid with theme entries or, in the case of theme-less puzzles, words or phrases that she really wants to use. Then she blocks out black squares and finally starts adding the rest of the fill.

Software can expedite the filling process by recommending words, drawn from a database, for various spots in the grid. “For instance, if you need an eight-letter word/phrase whose second letter is ‘U’ and last letter is ‘L,’ the software will do the heavy lifting,” writes constructor Elizabeth Gorski in an email. “It’ll immediately come up with: JURY POOL, HULA BOWL, BUZZ KILL, TUNA ROLL, PURE WOOL (to name a few).”

While an author could instruct a computer to “autofill” an entire grid, those who take their craft seriously scoff at the thought. “Relying on autofill as your primary ‘fill’ is like writing an article without proofreading, copyediting, or rewriting,” writes Gorski, who constructs for a variety of outlets, including her own, Crossword Nation. “An amateur relies on autofill.” Instead, a professional will carefully choose the best assortment of words that fit the grid.

But the phrases that the software suggests are only as good as the word database a crossword author maintains. A beginning constructor might kickstart her list with a community one—Crossword Compiler comes with a default list, for instance, which helped Steinberg get his start.

Serious constructors, however, religiously manicure personal lists of thousands of words, phrases, and names culled from every conceivable place—online dictionaries, movies, songs, conversations. “One time I bought a McGraw’s Dictionary of American Idioms and Phrasal Verbs, and I entered about a third of those [terms] into my word list,” says Joe Krozel, who’s been constructing for a decade, almost exclusively for The New York Times. “We go to all these extremes to build it out.”

Authors might also score the words in their lists so that the software can recognize which terms are most valuable and recommend those over others. “Every constructor has a slightly different system for scoring their lists,” says Steinberg, who uses a scale of 1-100, with “scores of 2 or 3 meaning, ‘I would hopefully never use this in a puzzle,’ and 100 being, ‘This is something I want to try and incorporate in my next puzzle.’”

Of course, what separates a good entry from a great one is partly a matter of opinion. But if there’s one thing about entries and their clues that cruciverbalists can agree on, it’s the less obscure or hackneyed, the better.

“Crosswords, like any art, should reflect life,” says Shortz, who started at the Times in 1993. “I’m looking for quality of the vocabulary, for words and phrases that are interesting, lively, [and] generally familiar, [with] as little obscurity and ‘crosswordese’ as possible.”

Modern phrases such as “SWEET TALK” or “DVDPLAYER” pass muster, says Shortz. But “ANOA”—which refers to a small water buffalo native to the Indonesian island of Sulawesi—hardly makes the cut. Neither does “ESNE,” a term for an Anglo-Saxon slave. Arcane words like these, as well as overused terms such as “Oreo”—compact at four letters, sociable with a few vowels—frequently slithered into crossword puzzles for decades as constructors, toiling with a pencil over paper, sought peacemakers between longer, more intractable entries.

But computers have facilitated a noticeable push toward less of this so-called crosswordese. Indeed, a conscientious constructor who maintains a tidy database has little excuse to depend on such lingo—she should have better words at her immediate disposal. “You could have justified it in a pre-computer environment,” says crossword solver Michael Sharp, who writes a popular blog called Rex Parker Does The NY Times Crossword Puzzle, but “it’s much harder to justify now.”

While solvers expect more in vocabulary, they also crave those “aha” moments that come with deciphering a tricky clue or discovering a theme. To that end, software can sort data (say, for instance, you have a theme idea where all the Ds are removed from words to make a pun, such as “rag racing” from “drag racing.” Software can delete the Ds from a word list, leaving mainly gibberish but some bona fide terms that a constructor can then draw from.) Clue databases can also make suggestions or tell constructors what’s been used before.

But dreaming up clever conceits is still by and large a human pursuit. “[Computers] can’t figure out what’s funny, [and] they can’t be subtle,” says Ben Tausig, the editor of the American Values Club crossword, an independent crossword that was once part of The Onion newspaper.

“The more complicated crosswords—those that rely on wordplay, multi-word phrases, puns, and a bit of trickery—are designed for the human brain,” adds Gorski. “They can’t be written by computers.”

And sometimes knowing how to make puzzles the old school way—by hand—improves the overall effort. “It’s not necessary, but it gives you the edge through cross-training,” writes Gorski. “Knowing the specific architecture of a puzzle (the rules of symmetry, word counts, etc.), in the long run, gives you a huge advantage.”

Yet, there’s one crossword gimmick that practically requires computer assistance: the wide-open grid—that is, a puzzle that’s nearly devoid of black squares and packed to the edges with long words.

“[Some] constructors have kind of made it a goal to build these Mount Everest crosswords that have the fewest possible words,” says Tausig. These puzzles rely on a “sort of post-human computational complexity,” he says—the human brain would have a hard time coming up with so many long, interlocking words.

Joe Krozel’s July 27, 2012, puzzle currently holds the New York Times‘ 15×15-grid record for fewest black squares, at 17. Krozel calls puzzles like this “paper tigers.” “They look really scary at the start, but then they start to fall away with a normal amount of work,” he says.

These extreme grids come at a price, however. Typically they contain few of the “Scrabble-y” letters—that is, the high-scoring Xs, Qs, Js, Vs, and Zs—that many solvers love, because it’s virtually impossible to stack so many words with uncommon letters on top of each other. Long words, “tend to contain a lot of those Wheel of Fortune letters—R, S, T, L, N, and the vowels,” admits Krozel, which “solvers tend to get a little tired of seeing.”

Sharp, the blogger, is one. “When it gets to the most extreme, you really are just trying to get words that will work,” he says. “So there’s not a lot of personality in these puzzles.”

Still, in the crossword world, “there’s a spectrum of solvers,” says Krozel, and “there are a certain number of solvers that like very sparse grids that give you very little opportunity to find a foothold.”

Anna Shechtman’s handwritten crosswords could be described as a kind of diary. References to music she’s currently listening to or movies she’s recently seen might inspire or wend their way into a puzzle, making it “a reflection of where I am today,” she says, like “some sort of really elaborate, kind of preposterous mood ring.”

Shechtman’s longer entries tend to be a bit livelier, “because it’s a human brain who has picked them out and tried to work around them,” says Matt Ginsberg, a computer scientist and constructor who created what he says was the first crossword constructing software back in the 1970s. (That proto-program didn’t go big time, although Ginsberg has had several other claims to fame, including a crossword-solving program called Dr.Fill. Tune in to SciFri on September 19, 2014, to hear more.)

Ginsberg, who’s currently overseeing a collaborative project to score the words in a 16-million-word list, is completely dependent on computers. “I have no idea how I would even begin to do this without computer assistance,” he says. “I don’t get how Anna does it.”

But he’s gaining some insight. He and Shechtman are working together on a crossword for the Times—Ginsberg is relying on software, and Shechtman on her pencil. “We’re good at different things, and we’re trying to sort of stumble into a way where we can each use our strengths and produce something better that we couldn’t have produced separately,” he says.

An early attempt hit an impasse when Shechtman provided Ginsberg with a partial fill, and his software revealed too many dissatisfying compromises to complete the grid. Now they’re working on another solution. When they find it, it’s bound to be an “aha” moment.