For this special “Class of 1984” issue of our magazine, we asked Kyle Mahowald, assistant professor of linguistics here at UT, to construct a crossword puzzle for us. In addition to being an accomplished linguist, Mahowald is also an accomplished crossword puzzle constructor, having placed 18 puzzles in The New York Times.
After you do the puzzle (which is roughly the equivalent of a Tuesday New York Times puzzle in terms of difficulty), read the interview I did with Mahowald about how he came to be a constructor, how he was surpassed by his husband (UT Austin assistant professor of philosophy Robbie Kubala) as a solver of crosswords, and how he went about constructing his puzzle for us, among other things.
Daniel Oppenheimer: What’s your backstory on crosswords? How did you get started, and how did you level up to where you’re publishing puzzles in The New York Times?
Kyle Mahowald: I started really doing them in high school, around 9th or 10th grade. The school librarian used to make copies of crosswords from various newspapers and put them out in the library, and so I just started doing them that way. Pretty soon I got really hooked on the New York Times sequence, where they get harder as the week goes on. At the beginning, I could do Mondays and Tuesdays, and Fridays and Saturdays were way past me, and then I got better over time, and got hooked on that cycle of it. At some point, I decided to try to make one for the school newspaper.
That was really fun, and so I got into constructing, as we call it. At the time there were these really active online crossword forums, and I started participating in those, and there were people in the forums who were themselves constructors and were super generous about mentoring new constructors. In particular there was a woman on the forum, Nancy Salomon, who was a long time New York Times constructor, and she had mentored a lot of people. And I reached out to her and said, ‘Here’s a puzzle idea.’ And she helped me turn it into a publishable puzzle, which ended up in the Times, and from there I was off and running.
Oppenheimer: How old were you?
Mahowald: I was 16.
Oppenheimer: Weren’t you at some point the youngest person in some category?
Mahowald: I was the youngest person to construct a Sunday puzzle, when I was 19, but only for a little while. That record was broken not too long after.
Oppenheimer: So how many puzzles have you done for the Times at this point?
Mahowald: 18. It’s been a couple of years since my last one.
Oppenheimer: You told me earlier that you weren’t solving Saturday puzzles until college, but you were placing puzzles in the Times when you were still in high school. It must be a very different skill, right?
Mahowald: Definitely. Constructing them is much more of a creative endeavor. You need to pay attention to language and notice words or phrases you come across in the world and think to yourself: Would this be fun to put in a puzzle? Would this be interesting to match with this answer? It requires an an aesthetic sense of what would make a fun, interesting puzzle. Solving them is more mechanical. It’s more about the knowledge know you have.
Oppenheimer: Are there tools one uses to construct puzzles?
Mahowald: The most widely used software is still the same one I used when I began constructing. It’s called Crossword Compiler. That’s the Windows-based one. There is another one for Mac called Crossfire. You plug in various answers you have, and they help you lay out the grid.
These days active constructors maintain word lists, with the words and phrase they want in their puzzles, and they also sometimes share them. That’s been a big shift since I started, the increase in widely available word lists, and a lot of the competition among constructors is rushing to put fun, interesting new words on their word lists and to be the first to use them in a puzzle.
Oppenheimer: Let me make sure I understand. A word list is a list of the answers, and people develop them independently of what the clues would be?
Mahowald: Right. You’re constructing the grid first, with the answers, and then the clues really are kind of an afterthought.
Oppenheimer: So you start with the words, and plug them into the grid.
Mahowald: If you have a themed puzzle, you start there, and typically the theme-related words get placed first. Then you fill the grid around that. And so you have a completed answer grid. And you want it to be the case that all the answers you put in the grid are interesting and aren’t so obscure that you can’t clue them. Then the clueing comes after that. When you submit a puzzle to someplace like the New York Times, the editorial staff is pretty happy to change as many clues as they have to. The decision to publish or not depends much more on the theme and the grid than whatever clues you submit, because if they don’t like a clue, they’ll just change it.
Oppenheimer: I want to pause on the word list a bit. What is it people are adding to their lists? Is it celebrities? Is it new slang on TikTok? Award-winning novels? TV show references?
Mahowald. Yes, all of that. The mix depends on the outlet. There are a lot of indie puzzles, for instance, that tend to skew towards the younger audience or toward a specific kind of niche audience, and so they don’t need to appeal to the widest possible demographic. For them, it may be the latest thing from TikTok and they’re confident that their solvers will know those references and be excited about them. The New York Times is more conservative. It tends to want to include something that seems as though it will have a longer shelf life.
Oppenheimer: How quick is the turnaround on a new cultural reference making it into a puzzle? If there’s a presidential debate, for instance, and there are some memorable lines, will a puzzle reference it the following week? Or is that a little too fast?
Mahowald: Depends on the outlet. The Times tends not to move that fast. Will Shortz used to say that it should have a shelf life of five to ten years. If in ten years someone is solving the puzzle, it should still make sense to them.
Oppenheimer: Am I right that you got your husband Robbie, who’s also on the faculty here at UT, into crosswords? And that now he’s better than you at solving them?
Mahowald: Yes and yes. He is definitely faster than me. He placed 21st at the last national tournament we went to, which is incredibly fast. He’s also written about them professionally. His research is in philosophy of art and aesthetics, and he published a really nice journal article about how crosswords can be thought of as an aesthetic form.
Oppenheimer: I wrote an article on that paper, which is how I found about your work as a constructor. Haven’t you also done academic work on crosswords in your own field of linguistics?
Mahowald: I’ve done a couple of things. I had a paper where we were looking to see if we could get large language model systems to solve cryptic crosswords, which are a really challenging type of crossword. That was a few years ago, and at the time the answer was no. It’s still no, but they’re closer than they were.
The other somewhat academic thing I’ve done on crosswords is an article for The Atlantic, with Scott AnderBois and Nicholas Tomlin, on the linguistics of crosswords.
Oppenheimer: That is a totally fascinating article. Everyone should read it, but while I have you here, tell me about green paint.
Mahowald: One way to think of crossword language is as a specialized form of English with its own set of conventions and its own grammar. It shares a lot of the same principles that we use to understand human language in linguistics, but it has its own quirks.
“Green paint” is often used as an example of a certain kind of answer that is totally legible by the conventions of English, but isn’t a good crossword answer, because it’s too arbitrary. In linguistics we have a concept we call “compositionality,” which is the way that in languages you can compose phrases out of individual words and even if the listener has never heard that combination of words before, they know what you mean. Even if you’ve never heard someone say “green paint” before, you know what I mean when I talk about green paint. But if it appeared in a crossword, people on the blogs would complain about it, because there’s not really anything to it besides green and paint.
Compare that to green tea, which would be a perfectly fine answer because even though it is compositional, in a sense, it is its own thing. Green paint isn’t any more special than blue paint or red paint. You want crossword answers to have their own meaning to them.
Oppenheimer. So Green Day, the band, could be a good answer.
Mahowald: Yep. That has nothing to do with green or day, really. You just have to know that it’s a band.
Oppenheimer: And even green paint could be a good answer, if the clue pointed you in the direction of the discussion that crossword lovers have had about green paint.
Mahowald. If it was clued as, say, ‘canonically bad crossword answer,’ then yes.
Oppenheimer: So are people using LLMs to solve crossword puzzles?
Mahowald: Yes, absolutely. One of my collaborators on that Atlantic piece, Nick Tomlin, works on this crossword solving software, Dr. Fill, which for years competed in the American Crossword Puzzle Tournament.
Every year this one guy, Matt Ginsberg, brought his laptop to the tournament to run the software, and it would try to automatically solve the crosswords. It was very fast in some ways, but it would usually get tripped up on some tricky theme. Ginsberg was convinced that Will Shortz, who was in charge of constructing the puzzle, was deliberately throwing curveballs to throw off Dr. Fill. One year, for instance, there was a tournament puzzle where every clue included a spoonerism, which means you reverse the initial letters of a phrase. So instead of birdwatcher, for instance, you’d have wordbotcher. In all other ways the puzzle was totally normal, but you had to realize that in every clue you had to do this operation. Then you could just solve it as a normal puzzle.
If you took the clues at face value, though, none of the phrases made any sense. They were all English words, but nothing fit together. So you had to make this leap, and Dr. Fill couldn’t do it.
At some point, Ginsberg teamed up with a team at Berkeley that was able to add neural language model techniques into Dr. Fill, and that seemed to push it over the edge to where it could win the tournament. As far as I know, it’s been retired since then.
Oppenheimer: Is it one of those things, like chess software, where once it surpasses human capacity, it will never be catchable again?
Mahowald: Mostly, although I would guess that you could still create themes that require a lot of flexible thinking that would be tough for any software to solve.
Oppenheimer: Let’s talk about the puzzle that you constructed for our magazine. My request to you was very general. I think I said that it should be 1984 themed, and that if possible, you should sneak in some UT related clues. Talk me through how you created it from those directives.
Mahowald: Is it okay to have spoilers?
Oppenheimer: Yes. Spoilers are welcome. If you haven’t done the puzzle yet, reader, go do it and then come back to this interview.
Mahowald: In a themed puzzle, you have to start with the theme, so I decided to begin with George Orwell’s novel 1984, and then take the phrase Big Brother from that. But the very crossword-y way of approaching that theme is to avoid the obvious concept, which would be about surveillance or tyranny or something in that vicinity, and ask: what else can I do with that? The idea I came up with was to take some celebrities who were relevant in 1984 and rather than cluing them as themselves, clue them in reference to their actual big brother. So that’s the theme here.
The clue is the first name of the big brother, and his birth year. So, for example, the clue is Michael, born in 1949, who is a musician and faculty member at Lehman college. It’s Michael Bacon, Kevin’s older brother, and the answer is Kevin Bacon. One of the clues is Jackie, Tito, Jermaine, and Marlon, and I give birth years for all of them, and so the answer that you need for that one is Michael Jackson.
Oppenheimer: And you have Madonna’s older brother?
Mahowald: Yes, Anthony Ciccone. Ideally you want clues which, if they’re not gettable right away, they make sense when people solve them. There’s no real reason to know that Anthony Ciccone is Madonna’s brother, but maybe somewhere rattling around in people’s head is the idea that Ciccone is Madonna’s original last name. And so once you have a few letters, you’re like, oh, right, that’s Madonna. For other clues the psychological reward is a different kind of thing. With the clue Gary Garland, born in 1957, former Denver Nuggets basketball player, he turns out to be Whitney Houston’s older brother. Most people don’t know that her brother played in the NBA, but it’s kind of an interesting fact once you fill it in. That’s the reward.
Oppenheimer: That reminds me of Robbie’s paper on the aesthetics of crosswords. There’s a little kind of aesthetic tickle in your brain when the answer is revealed. Whitney Houston’s brother was a basketball player; I didn’t know that. Or Kevin Bacon’s brother is a musician, and then you remember that they have a band called the Bacon Brothers, and that’s cool.
Mahowald: That’s exactly why I did it. For all of them, I wanted it to be an interesting answer.
Oppenheimer: In addition to the Big Brother themed clues and answers, there were a fair number of clues that had some kind of 1984 aspect to them, which was also nice.
Mahowald: Yes, and it’s a bit unusual. Most of the time when you have a New York Times themed puzzle, you just have the theme and then the rest of the clues are unrelated to that theme. But because this was a special edition puzzle, I tried to pack in as much 1984 stuff as I could.
Oppenheimer: I forgot to tell you, we had to correct you on one clue.
Mahowald: Oh yeah?
Oppenheimer: You had 38 down clued as the subject of the film Dunkirk, but the answer that works in the puzzle is D-Day, and Dunkirk was not about D-Day. So we found a film about D-Day, The Longest Day, and subbed that in.
Mahowald: Good catch.