Puzzle Trouble: Women and Crosswords in the Age of Autofill

Puzzle by Anna Shectman, The New York Times, 5/29/14
Puzzle by Anna Shectman, The New York Times, 5/29/14

In March I attended the American Crossword Puzzle Tournament, an annual gathering of speed-solvers, puzzle constructors, and other eccentrics hosted by Will Shortz at the Brooklyn Marriott. My responsibilities at the event were threefold: I was there as a judge, as Will Shortz’s assistant, and as a constructor with a puzzle in the tournament. At the time, I had just graduated from college and was working four days a week for Shortz, helping him edit puzzles and prepare them for print in the New York Times—by many measures, the gold standard of grids. The logistics of the tournament, which, when articulated sound like the start of a bad joke (“How do you fit 700 crossword solvers into a conference room?”), had occupied us for months. This was my first ACPT—my crossword cotillion—and I spent the first night’s mixer matching faces to bylines.

Since I began creating puzzles at age fifteen, I knew that as a humanities-minded female I was an outlier in the CrossWorld, a realm dominated by middle-aged men hailing from the natural and computer sciences. But truthfully, I am an outlier among other outliers: you have to be a little touched to find creative inspiration in a 15-by-15 black and white grid.  And the puzzle mishpucha has a strong pull towards camaraderie—strong enough to undermine any sense of my own difference at the ACPT.

Crossword culture has been male-dominated for years. But it wasn’t until the event’s second night that the topic of crosswords’ “gender problem” emerged from the mouth of its most unlikely messenger: David Steinberg, a seventeen-year-old puzzle wunderkind. Steinberg’s first crossword was published in the Times when he was fourteen. He has since had twenty-eight puzzles appear in the Times and has many more slated for print. In 2012, he founded the “Pre-Shortzian Puzzle Project,” which aims to digitize every New York Times crossword since the puzzle’s debut in 1942 (all “Shortzian” puzzles, from 1993 onward, are already digitally available). Part of Steinberg’s project involves identifying formerly anonymous constructors and charting changes in constructor demographics over its seventy-two-year Times history. A natural data-head, Steinberg lives in Rancho Palos Verdes, California and seems destined for Silicon Valley after college.

On the second night of the tournament, Steinberg was given the podium to brief us on the “Pre-Shortzian Puzzle Project.” The later half of his speech focused on women in crosswords: “XXword constructors,” as they’re never called. Steinberg presented his data on the role of gender in the authorship of Times crossword puzzles—a statistical analysis that he developed for his high school science fair.

To the disappointment, but not the surprise, of many in the audience, Steinberg revealed that the gender disparity in crosswords has steadily worsened in the past two decades. Under the two editors before Shortz—Will Weng (1969–1977) and Eugene Maleska (1977–1993)—women constructed 35% of all puzzles, and in the reigning “Shortz era,” women account for 19% of all puzzles. Moreover, most of the female-made grids in Shortz’s time have appeared on Mondays or Tuesdays, the week’s easiest puzzles according to the Times’ sliding difficulty scale (the “M-T ghetto,” as they have been crassly called by one puzzle blogger). It’s actually quite difficult to construct an “easy” but engaging Monday or Tuesday grid, but that women appear to be ghettoized in the early, less glamorous days of the week is no doubt troubling. 

At this point in his speech, Steinberg suggested that the ever-widening gender gap in crosswords might be explained (or “boysplained,” as one puzzle blogger snarked) by the field’s shift to software-facilitated constructions. Like photography, film, and any number of once-analog trades, the process of puzzle-making has been digitized in the past twenty years. With few exceptions, and I count myself among them, constructors use programs like Crossword Compiler or Crossfire to assemble their grids.

Steinberg’s statistical analysis offered cogent data to support his hypothesis that the rise in crossword software is correlated to the widening gender gap among constructors. As puzzle-making became increasingly informed by programming, it began to replicate the gender imbalance of tech culture (and “nerd culture” in general). To support his findings, Steinberg invoked the “numerous studies that suggest that females are far less likely than males to enter computer and other technology-related fields.”

The changes that software has made to the puzzle game starting in the mid-1990s can’t be overstated. Almost every constructor I know has forsaken graph paper and pencils without much hesitation—and certainly without remorse.  Crossword Compiler and Crossfire simply make better puzzles faster than anyone could possibly make by hand.  The programs allow constructors to build (or import) word databases, which the software organizes into an interlocking grid.  This is not to say that the software itself merits authorship of the puzzle. A software-based grid is not immaculately conceived: the constructor is still responsible for the puzzle’s theme, its clues, and for fine-tuning the puzzle’s fill. Often the software will alert the constructor to a quadrant of a grid that is simply “unfillable” by its own algorithm. In this scenario, a constructor can change the placement of black squares and ask the software to try again. Or she can try to outsmart the software and fill the corner “by hand”—the old-fashioned way.

In the age of software-based construction, the puzzle-maker’s obsessive energy, once focused on interlocking words in the grid, has been displaced onto building a word database. Small boxes cede to Big Data. I overheard many conversations between constructors at the ACPT about how frequently they were updating their word lists. Updating, in these terms, means adding words and neologisms to one’s database and redefining the software’s construction priorities: instructing it to privilege some words in the list over others. Through a series of not terribly technical operations, the constructor can tell the software that the obscure and overused ETUI (“needle case”) should only be used in desperation, whereas the newly-added BOYSPLAIN should be given top priority in a grid. (The chances that Shortz would accept a puzzle with BOYSPLAIN are slim to none.)

To use the software artfully, one has to tend to it—constantly updating and re-ranking the words in one’s database to reflect the evolution of language and common usage. It’s hardly the project I signed up for when I began constructing. Every move to digital has its naysayers and analog-fetishists, and in other mediums it’s not a type with which I identify.  I’m no Luddite, and yet I continue to construct by hand. I like that the frustration of my process is reflected in my deadened eraser. I view the 15-by-15 grid as I do any other written medium, in which word choice is just that: a choice, and not the byproduct of an algorithm (even if it is an algorithm I’ve manipulated). In short, I didn’t sign up for algorithms.

In this sense, I completely reinforce Steinberg’s slightly tone-deaf claim at the close of his ACPT speech that women may have been “left behind” when the dominant form of constructing became “less of a literary exercise and … more of a mathematical exercise.” Flirting with Larry Summers-like essentializing based on gender, Steinberg was at the very least guilty of generalizing. Two of the most exciting additions to the crossword scene, for example, are Aimee Lucido and Zoe Wheeler. Aimee is a software engineer at Facebook and Zoe is beginning her PhD in computer science this fall. Needless to say, they both construct using software. 

Ultimately, Steinberg’s speech did more than reveal him to be a budding statistician. It also shifted the conversation and the burden of crosswords’ gender imbalance away from male puzzle editors. For the past twenty years, the CrossWorld has been reigned by a puzzle androcracy: Shortz at the Times, Rich Norris at the Los Angeles Times, Mike Shenk at the Wall Street Journal, Stan Newman at Newsday, and Tim Parker at USA Today. Last summer, Ben Tausig, editor of the phenomenal AV Club crossword, wrote an essay for the Hairpin about women in crosswords, positing an alternate hypothesis for the widening gender gap: implicit byline bias. While no one is accusing these men of motivated discrimination, Tausig suggests that the dearth of female puzzle editors informs the decline in female constructors.

A successful puzzle is an index of its maker, reflecting her interests, humor, and use of language. In his essay, Tausig proposes that like attracts like—that male editors who profess gender-blindness might gravitate nonetheless towards male-made grids, rife with baseball trivia, Tolkien beasts, and car parts (to traffic momentarily in the field’s shallowest stereotypes).

As for the Times, Shortz is fully aware that his rubric for acceptance is subjective, and he makes a point to surround himself with varied puzzle perspectives. Two of his three test solvers are female, and he is known to seek out young (and, in my case, female) assistants to remedy the blind spots in his knowledge base. When I was twenty, I submitted a puzzle that he rejected. He cited MALE GAZE among the entries he found unworthy of publication. I don’t doubt that a woman or a younger editor might have deemed that entry an asset as opposed to a demerit. I also don’t doubt that if my puzzle had been stronger overall, he would have run it despite his unfamiliarity with feminist film theory.

Tausig speculates, “If women perceive an editorial bias they may never even reach the point of submitting their work.” As an editor, he has gone out of his way to solicit contributions from female constructors and to mentor them in grid design and construction. He sees the problem as generational, one that a new crop of prominent female puzzlers could rectify.

He and Steinberg are not alone in speculating about the lack of women in the field. Many of the female constructors I’ve spoken to posit their own theories for their underrepresentation, all as persuasive—and as problematic—as those of Tausig and Steinberg.

In a Facebook message studded with ALL CAPS exasperation, Aimee Lucido offered a variety of alternate hypotheses, including the possibility that some women are turned off by the “bitchiness” of crossword culture. She was referring specifically to crossword blog culture, whose titans Amy Reynaldo (“Diary of a Crossword Fiend”) and Michael Sharp (“Rex Parker Does the NYT Crossword Puzzle”) are known for their sharp elbows and unsparing puzzle reviews. “Maybe men see negative feedback as a challenge to do better and women see it as a warning to stay away,” she wrote. Despite the bold political incorrectness of Aimee’s theory, I’m woman enough to admit that a Michael Sharp review has totally made me cry.

When I spoke with Amy Reynaldo, who also co-edits the Daily Celebrity Crossword, she invoked the structural changes in women’s lifestyles that may also account for the decline in female constructors: “Women who were constructing in the ’50s and ’60s were doing it to keep their brains alive while home with the kids.”  She suggested that working women, many of whom are still responsible for the bulk of housekeeping and childcare, don’t have time to create crossword puzzles. 

Tracy Bennett, a New York Times crossword constructor and editor of the puzzles in the women’s magazine BUST, echoed Reynaldo’s thoughts when describing her own balance of work, childcare, and constructing. She spoke candidly about the jealousy she once felt for constructors with fewer professional or personal responsibilities and more time to hone their craft. In an email she wrote, “Not only do I work full-time, but I parent full-time as well.  So I have to make excellent puzzles (still working on that), because I can’t make that many.  And if I’m making fewer puzzles, I’m getting feedback from the editors at a slower rate, and my learning curve may take more time because of it… I’ve accepted that I’m a tortoise in this race.” 

A “pastime of privilege,” puzzle-making requires very specialized skills and offers very little compensation. In this sense, it’s remarkably well suited to the brogrammer culture skewered on shows like HBO’s Silicon Valley—spaces buzzing with mental agility and free-floating virginity. It’s not that women aren’t up to the challenge of tech-based constructing (Bennett and Reynaldo also use software to make crosswords), but the decline in female puzzle-makers may be a symptom of the aesthetics of tech culture, not the technology itself.

After Steinberg’s talk at the ACPT, puzzle-maker and software guru Matt Ginsberg took to the stage to discuss Dr. Fill, a computer program he developed that can complete crosswords with startling speed and accuracy. In a 2011 article in the Journal of Artificial Intelligence Research, Ginsberg suggested that Dr. Fill is among the top fifty crossword solvers in the world. Like I.B.M’s Watson or computer chess, Dr. Fill is an A.I. experiment: an opportunity to see if computers can “do poorly what humans can do well,” as Ginsberg likes to say. Using Big Data (which in this case means databases of previously used clues and Google search histories), Dr. Fill’s algorithm doesn’t attempt to solve a crossword puzzles so much as it attempts to maximize the probability that a puzzle’s “fill” is correct.

But Ginsberg is humble about his software’s limits: Dr. Fill often gets stumped on crosswords with inventive themes and new vocabulary—which is to say, it’s stumped by the human imagination.

Theoretically, software using similar resources as those employed by Dr. Fill to determine a word’s “puzzle-worthiness” could be used to construct a crossword without human intervention. The fill could be pooled from Google and Twitter word trends, while its clues could be drawn from old puzzles and dictionary definitions. This hypothetical franken-puzzle would eliminate the question of gender in puzzle authorship—along with the question of authorship itself—but it wouldn’t be nearly as fun to solve as a Steinberg- or a Lucido- or a Bennett-made construction. Evidently, in the field of “cruciverbalism,” humans still have the game on lock.

I was fourteen when I discovered that crossword puzzles have authors—that they aren’t computer-generated, but rather are painstakingly and symbiotically crafted by a constructor and an editor. It took me a year or two of dedicated solving to develop my own auteur theory for crosswords, noticing that a Liz Gorski construction is often marked by a visual element embedded in the grid, whereas a Martin Ashwood-Smith grid usually contains large stacks of fifteen-letter entries. Finding the person in the puzzle was part of its fun. Like all auteur theories, mine developed in admiration of the quirks, artistic flairs, and signs of difference among constructors in the CrossWorld—even if it is an increasingly homogenous group by any statistical measure.

I’ve come to especially cherish a David Steinberg byline and the uncanny access it provides to his worldview (if not to his database). Recent Steinberg entries in Times puzzles include BIKER CHICK, HOT DATES, FILE NAME, SOULJA BOY TELL EM, and GENDER BIAS. I imagine that he constructed the grid with this last entry around the time of the ACPT.

Our September issue (Vol. 2 No. 3) will feature an original crossword by Anna Shechtman. Subscribe to the American Reader in our Shoppe.