Girl here.
I'm in yon fogarty and today i'm here with natan last, author of a fabulous new book about crossword puzzles called across the universe.
At 16, natan was the youngest person ever at the time to get a sunday puzzle accepted by the new york times, and now he teaches crossword writing classes and regularly contributes puzzles to the new york times and the new yorker.
Natan, welcome to the grammar girl podcast.
Thank you so much for having me.
It's a thrill to be on.
Oh, I'm excited to have you.
I really loved your book.
I think it's going to be, I'm going to get some as gifts this holiday for people.
I learned so much.
So you're going to hear me say I was surprised so many times in this podcast.
So one, I was surprised to learn that American and British puzzle conventions are different.
They're very different.
Yeah, they kind of play to type a little bit.
So the British style are pure cleverness.
In look, they are lattice-shaped, so they look like waffles.
And they don't have as many crossings.
There are fewer crossing words.
But each clue is kind of built up out of puns and riddles and sly bits of language.
Whereas the American style that we're all familiar with has that huge swatches of white space and more pop culture, cleverness here and there, but a lot of trivia as well.
So if I went to do a British crossword puzzle would I find it.
Is it hard to cross the two styles or do people go back and forth pretty easily?
There are definitely bilingualists for sure.
It really, really scratches a different itch and uses a different part of the brain.
The kinds of clues you would see in a Friday or Saturday New York Times crossword, sometimes with a question mark indicating that wordplay is afoot.
Every single British style cryptic crossword they're called is like that.
And so if you like that style of the Friday and Saturday deviousness, then the British style is definitely for you.
And is one style older than the other?
Yeah, the American style crossword sort of is the one that comes first.
And it's not really until after World War II where, for instance, the Times starts publishing a crossword that the divergence happens.
But as early as the 1920s and 30s.
The two styles are pretty much, or the two countries, I should say, are pretty much doing the same thing.
Crosswords kind of like we know them today when we open up New York Times.
Okay.
So the other thing that surprised me is I was just stunned at how long it can take from when a crossword puzzle is accepted to run in the New York Times between like, when it actually shows up in the paper or on the website.
It can really be a long slog.
Yeah.
When I was younger.
I remember you know you get so excited when you get that coveted crosswords yes, subject line from Will Shorts and his team.
And it can be a year, sometimes more, until you see publication.
One of the interesting things that happened is that a huge crop of blog puzzles and this indie scene where people wanted to respond to the moment right
Wanted to use crosswords more as a barometer of language, to use it to kind of take the temperature of the culture and the words being spoken in that culture.
And so I remember solving a Brendan Emmett Quigley who's a great constructor puzzle in 2009, the day after Sonia Sotomayor had been confirmed.
And seeing her name in the grid kind of had a hand in confirming that this was indeed common knowledge, that this was a name.
We all should know that, rather than waiting for some other codification, right away a crossword constructor could lift something up into the realm of common knowledge.
Yeah.
Yeah, I love doing puzzles that have current events in them.
Why does it take so long at The Times?
It's mostly just due to supply and demand.
There are so many really good puzzles that they get.
That's only increased as The Times games app has expanded.
It's got more eyes on it.
The interest in construction has expanded.
That started to happen right when The Times went digital before COVID.
And in COVID a ton of people got really really, really excited about trying their hand finally at construction.
These are longtime solvers who really love the game.
And so there's just a lot of good puzzles.
It takes a long time to read each submission individually and figure out where it should go, if it's times worthy.
And yeah, it's just it's just there's an explosion of interest.
And so there can only be so many puzzles a week.
Yeah, yeah.
And one reason I expect you to know this actually is that you actually interned with Will Short.
So what was that like?
And like, like, how did he change the game?
That was a ton of fun.
I was Will Schwartz's intern when I was at the end of my high school and early college days.
And I would just take the train up.
I grew up in Midwood in Brooklyn.
So I'd take the Q to the 4, to the Metro North, to his wonderful hamlet called Pleasantville where he worked from home.
His first adult I met who worked from home.
And he'd greet me in crossword print PJs at like 1030.
And I was like, this is the life.
This is how you do it.
And it was wonderful.
I mean, we... it was a very small operation then.
So we would go up, we'd print out that day's puzzles to be edited, put them on a clipboard and just go, clue by clue, look things up.
If we had to, I would Google things while he looked things up in books, and I would always win that race early, early moment of that divergence.
But it was really nice.
I mean, I would also get experimented on.
So he'd do the NPR weekly puzzle and, He'd test that out on me before we went up to edit.
And if I got him too quickly, he'd always make it harder.
I remember being the kind of patient zero on the really kooky and wonderful Sunday NPR puzzles that he runs.
That's amazing.
So I was also surprised that... So you were doing more than fact-checking.
You're changing some of the clues on people's puzzles, right?
Yeah.
For a long time it was sort of an insider secret that Will Shorts and editors in general can change up to 90 of the clues.
Wow.
Which is a lot, right?
And so I mean, for a long time it was kind of unnerving to wake up on publication day, because there was an era where you didn't get any proofs.
So you'd wake up and you'd see your clues massaged into an entirely new direction by a firm hand, but maybe not in a way that you necessarily agreed with.
Sometimes that's just changing difficulty and tweaking the trivia.
That's asked for tweaking a reference.
Sometimes it's alighting on a really fun question mark or devious clue that can kind of spice up an existing puzzle.
Other times it's simple things like the New York Times clue.
Syntax is a structure and not everyone immediately knows how to imitate that structure.
So it really can be anything from making it harder to just conforming to time style.
And yeah, older salts who have been making puzzles for a while.
We'd only change maybe 5 10, 15 of their clues.
But Often, if someone had never made a puzzle before, the theme was really fun.
The answers in the grid were totally interesting, but the clues in particular needed that kind of TLC.
Wow, that's fascinating.
Okay, and so I want to go back to the crossword puzzle pajamas because that's adorable.
So, like, is his whole house, like, crossword kitsch?
Yeah.
Totally.
Yeah.
Will Schwartz is a real his house is a museum and he's a real connoisseur and collector.
One of my jobs when I was a high school student was to go into his extremely haunted basement and page through.
He had hard copies of 1925 New York Herald Tribune.
And in the mid 20s, there is a crossword craze.
So the first ever crossword puzzle book comes out in 1924.
And then, in an echo or pre-echo of those crossword print PJs, people are going to crossword balls and wearing black and white checkered dresses.
There's this wonderful crossword dress with a peplum cube hat.
And everyone has been bitten by the crossword bug.
It's called crossworditis.
The sort of generation above treats it like a disease, a madness, a craze.
And so Will Schwartz's house is really interesting. an homage to ephemera from that time.
He has, you know, crossword bracelets that have little five by five grids on them.
He has, you know, puzzles from that era and everything from you know.
The clock in his study is a crossword and the minute and second hands are pencils, right.
It's all crosswords.
That's amazing.
So, I mean, what were you going into the basement for though?
So I was combing through pages from the 1925 Herald Tribune to try to find any information, cultural residue from the crossword craze.
And you know four pages in I'd find an advertisement for crossword print shoes that had black and white buckles.
I'd find 20 pages later an invitation to a crossword ball in which you could see sort of dapper Dans with their pork pie hats.
And again, these sort of 1920s flappers in black and white dresses.
So I was looking for evidence like that, because wool shorts at the time was, I think, combining a lot of this into some kind of authoritative compendium.
But it was mostly just an education in, you know, the jazz age, right?
And what was going on in the newspaper world at that time.
How fun.
And this is a complete aside, but I had no idea that he was also a huge table tennis player.
Will Schwartz is really into table tennis.
He purchased a table tennis club while I was his intern.
I remember I was so young I didn't really understand that you could amass that much in savings from you know, sudoku and crossword puzzle books and then essentially liquidate all of it so that you could play more games.
In this case by a table tennis club in pleasantville and he is, i think, essentially the cal ripken jr of table tennis, where he will sort of forego travel and events because his streak of playing table tennis every day cannot be interrupted.
And yeah, he's a very talented player.
Uh i i, you know, I think of myself as someone who's pretty good at sports and love table tennis, and he just whooped my butt every time.
He's great at it.
That's amazing.
It's like two divergent things he's really, really good at.
That's wild.
I aspire to be that good at two completely different things.
Me too.
So the mathematical analysis of the games also just fascinates me.
You talked in the book about the investigations into which words are crossword-ese.
Yeah, totally.
So I think longtime solvers have all encountered crossword ease, which are those often vowel heavy words that, because English is so so, so consonant heavy thanks to its Germanic roots, you need that release valve of.
You know we all speak crossword French, right?
Like uh uh, and we all those vowel heavy words that go underneath, these consonant heavy words that come from Germanic.
And I think words like area, uh And era are extreme crosswordees in the sense that they appear in crosswords.
So so so, so much.
And we get kind of bored as solvers and as constructors having to clue them for the umpteenth time.
But there are also words that are crosswordy.
And this software engineer, Noah Veltman, took words that appear a ton in crosswords but then don't appear in books, in this case using the Google Books corpus.
And those words are words like ursa.
Or Captain Smee from Peter Pan, where the only context in which you encounter that letter combination is in crosswords right.
So those are the crosswordiest words.
And there's like numerical rankings that constructors can look at, right?
And say, how crosswordy is this word?
Yeah, exactly.
The field has become extremely soaked in quantitative data for good and for ill, I think.
But you know, there's a great website called xwordinfocom in which you can analyze a puzzle right.
You can see which words have never appeared in a crossword before.
So they're exciting because they're debuts as far as the New York Times is concerned.
Or you can see which words have high Scrabble value, right?
Which is to say if you tallied up the letters rankings in Scrabble because you've got a lot of Qs, Xs and Zs in a word, a word like jacuzzi scores really high in Scrabble value and is an interesting word but also interesting.
Getting those J's and Z's into a puzzle can often lead to interesting words going down as well.
So there's tons and tons of data for the curious, for sure.
Yeah.
So do people who write, I keep wanting to call them writers, but are they constructors?
Or what is the right word for people who make crossword puzzles?
Yeah.
Constructor, I think, is the sort of modal term.
But I think that one of the things I talk about in the book is that I really think that different words that people use have a different crosswording ethic underneath them.
So I think, I mean, people who love to do the cluing, right?
Writing clues is like writing.
And so I think a lot of people who are on the literary side and maybe favor the clue part of the process call themselves writers.
Building a grid is extremely mathy.
The writer, George Parekh, you know, referred to the like arithmetic of letters, right?
And it found that part very frustrating where all that matters is that words are a certain length and that they're, you know, they're digraphs.
So one letter next to another fits nicely.
And that, part, I think, lends itself to engineering and math types.
And people who favor that aspect tend to call themselves constructors.
People call themselves creators.
Writing a good theme is kind of like planning a really fun scavenger hunt that only takes 30 minutes.
And that feels like creation.
That feels like a creative act that's different than writing clues and filling a grid.
And so there's all these words, but I would say constructor is the one that people use the most.
And then, if you are ever feeling kind of playfully snooty, there's cruciverbalist uh, which I I, I almost never say out loud because it takes too long to say, it's hard to you know.
I think I write in the book that like that, that that's the kind of word that you can kind of feel its glasses falling down on its nose as you say it.
And so it's a wonderful word, but constructor is, uh, is the term of art, I think.
So do one group of these people or the other get more excited about using a word that has never appeared before?
And actually, if you use a word that has never appeared before in the New York Times, for example, does it make it more likely that your puzzle will be accepted, since it's so competitive?
Yeah, that's a great question.
People definitely strive to debut a word.
That is kind of more for cross-world bragging rights than anything else.
I don't believe at least it wasn't the case when I was editing with Will that you would look at a word and say has this ever run in The Times before?
And if it hadn't, bump it up.
People tend to look at a crossword grid, and Will would do this too.
You'd put checkmarks next to words you thought were interesting.
A word like jacuzzi would maybe get a checkmark.
If you didn't know, a cultural reference Will might put like a question mark or even a negative sign next to it.
Right.
You're almost quantitatively assessing these words, but you're still using gut judgment.
You're not going to the database.
It's definitely the case that when people submit a puzzle, they like to dot it with words that have never appeared in a puzzle before.
And I think the editors can sense when a puzzle is fresh.
A puzzle has words that are new either because they're newly in the language.
One of the lovely things about puzzles is they do kind of hold references. this like somewhat delayed mirror up to our own language or because they're, you know, new proper nouns, right?
Whenever like a new Star Wars character with a weirdly spelled name comes out, everyone rushes to put it in their crossword, right.
Awesome.
You talked about the databases and I do want to talk about how computers have changed crossword construction.
But first let's talk about, like how crosswords were made before computers.
Yeah, a lot of blood, sweat, and tears.
My dad is a public school math teacher, and so scratch paper in my house was graph paper.
And so I would just use graph paper and Google for the most part.
Earlier than my time, people would use this thing called a crossword puzzle answer book, which is this enormous over 1000-page book.
And it looks like a dictionary, but the headings aren't like A to B, B to C.
Instead, the headings are blank blank, blank N, blank D, And so it's all words that fit that pattern right.
There are seven or so letters or whatever.
There's an N in the fifth position and then D in the seventh.
And then the next page would be blank, blank, N, blank, blank, blank, D or whatever.
And so just to find letters that fit a certain string, required this big book in combination with the big brain right.
With just thinking about it.
And so for a long time, it was a lot of graph paper, pencil, Lots of erasure.
You know, I checked on eBay and you can get the crossword answer book for just 7 with free shipping now.
It's like 1200 plus pages.
That's amazing.
It is amazing.
And there's a lot of them.
So because I think because now everyone's using computers.
So let's talk about when people started using computers to make crossword puzzles and like how long did it take before they could get puzzles they could sell?
Like, I'm sure it was really bad at first, but like, you know, how did it end up?
Yeah, the first chapter of the book profiles Eric Albert, who is the first person to really use software to try and create the grids, to build puzzles.
And Eric really wanted, he was a software engineer for a long time.
He was getting a little bit antsy to try something new.
And he was like, maybe I'll be a full-time crossword constructor.
And everyone was like, it'll never work.
It takes too long to make them.
It doesn't pay that much.
And you're better off just doing some other programming job.
And he said I wonder if a computer can help and built a rudimentary program that helps fill a crossword grid.
And the key innovation is that it uses a ranking system.
So actually Eric ended up spending something like three years going through every single word he had in this big, big dictionary file, right?
750,000 or so words and saying, is this a good word or not?
I'm ranking it on a scale from... yucky to fabulous.
And the main thing the crossword software did is privilege the fabulous over the yucky which is, to you know, filling the grid is tricky enough on its own, but not super complicated.
What's hard is filling a grid well, filling it with words that are interesting.
Filling it with words that are interesting because Eric likes them.
And a fun thing happens in this era where people start using words.
This item called a word list, where they, you know, have a long dictionary file, and it comes with a bunch of rankings.
Abbreviations are smaller numbers.
Fun phrases like you know, free jazz or whatever that have J's and Z's and our and our phrases and fresh have higher numbers.
And then people spend a lot of their time manicuring and tweaking that list, putting putting stuff in that only they know would put in a crossword, and taking stuff out that they happen to think is not particularly interesting and wouldn't be a word or a phrase that they would enjoy seeing in their own puzzle.
Yeah.
So I imagine, like you wouldn't want I mean in the very beginning probably there were prepositions like of, or articles like and like you don't want those in a puzzle.
Yeah, those are really boring.
Crosswords have all these conventions, so there's only words of like three or more.
And exactly as you're saying, it's because it's impossible to find words, an interesting clue for the word of right.
These function words, they're just, just not that exciting on their own.
You know, the, the, yeah, you're going to demote words like that.
You're going to take things out that feel obscure.
You're going to take abbreviations out.
People feel differently about you know scientific terms, words that aren't in English and everyone's going to.
In this weird way, the software ends up letting folks express their own taste a little bit better, right?
Because they're making these thousands of aesthetic micro decisions about, but goodness of words.
Yeah.
And so did Eric Albert, he did end up selling puzzles, right?
Yeah.
He starts this journey in 1989.
And in the early 90s, he's a full-time crossword constructor for some years.
He did.
Yeah, he did it.
It worked.
And he submits his first grids and they get accepted by the Times.
And what's really interesting Interesting is that Will Schwartz has no idea.
The grids and the answers in them are good and clean, right?
As a constructor might say that there's not a lot of obscurities.
There's not too many abbreviations.
And at that time, no one could tell that these were, you know, in large part computer generated.
Did he feel tricked when he found out?
I think that he was okay with it, because the thing about crosswords is that that kind of arithmetical part of putting the words in the grid has always been interesting, susceptible to automation and has always had computer assistance in the offing.
But the clues were all written by Eric.
The themes were all brainstormed and devised by Eric.
And so there's still all of this human ingenuity shining through.
Yeah.
Great.
And then later a guy named Ginsberg I love.
He came up with a program he called Dr Phil F-I-L-L, which I imagine is like filling in the grid.
And his whole thing, it seemed like a It was a journey.
Yeah, he's really funny.
So Dr Ginsburg, Dr Matt Ginsburg is, you know, he's like this storied artificial intelligence researcher.
He has like a doctorate from Oxford in math at the age of 24 I think and starts a bunch of AI companies and starts building crosswords.
He becomes a constructor and then he goes to the first crossword tournament where people solve as quickly as possible.
It's a speed solving tournament.
And people are so fast that he gets mad.
He's kind of out, he's out for revenge.
And he, he wonders if he can build an AI that solves puzzles.
So you know we've talked about computer assistance, aids in building a puzzle.
But here he's like I wonder if I can, if I can do this.
And there the community is a bit has its eyebrows raised for sure, because it's one thing to interlock words with other words, that feels confusing, computational kind of from the outset.
But writing good puns, coming up with, you know, clever misdirections, brainstorming a theme.
They feel very human.
And in fact at the very first tournament that Dr Phil competes in, Will Shorts, who's the organizer of the big tournament, has 150 buttons made that say you know, I beat Dr Phil.
Very much team human at the moment.
And Dr. Phil actually underperforms And Will Shorts gets it exactly right.
I think he handed out 143 buttons or something like that.
And all throughout this time, Matt Ginsburg is wondering, how can I make this software better?
If you fast forward 10 years to the pandemic, he ends up teaming up with a Berkeley natural language processing and AI shop a bunch of researchers working who in their spare time had gotten into puzzles and were trying to recreate Dr Phil.
And they ended up marrying what we think of as kind of modern machine learning techniques, basically just screaming through just tons and tons and tons of data with the kind of hard-coded knowledge about crosswords that Dr Phil had in its silicon brain.
So it knows about anagrams, right?
So if you have like a pun like the satanic serves, which is a pun on the Russian novel The Satanic Verses,
But now it sounds like it's about really devilish tennis serves.
It doesn't know that that's a phrase because no one's ever said that phrase before.
But it understands that it's composed of a bunch of words.
And it understands that if you rearrange the letters in one of those words, it's the whole new phrase is something it's seen before.
So it kind of does know a little bit about a typical crossword theme.
And that tournament in 2021, I believe, where Dr Phil wins.
You know it beats everyone else, but it's not sort of allowed to compete uh, in the tournament, because the tournament is for humans, right?
Yeah, it ends up winning.
It ends up being everyone.
It beat everyone.
Oh wow, i missed that part yeah, and so people were probably pretty annoyed.
People were annoyed.
I think that um, matt ginsburg did the right thing.
You know, i interviewed him.
He said he talked to all the other game solving ai people.
So the person who made the, The AI that beat the first time this happens is in Backgammon in the late 70s.
And, you know, of course, ever since then, AI game solvers have plowed through chess.
They've plowed through Go.
They've plowed through Texas Hold'em.
And there is a sense with language that we're very protective.
I mean, this should be our thing.
We're humans.
We're funny.
We're creative.
You know, we're infinitely productive when it comes to words.
And all of the other game solving AI creators were like, you should hang up your cleats, you know.
If you win twice in a row, people really start to think you're rubbing it in their faces.
If you win and then lose, everyone calls the first time a fluke.
You should quit while you're ahead.
And so he does.
And since then, we haven't heard much of crossword solving AIs.
But it is kind of fun that, after a decade of not winning and after a decade of, in particular, human crossword makers deliberately trying to trip up the machine and succeeding, Dr Phil ends up winning in that 2021 tournament.
And then calls it a day.
He's hanging up his chips.
Yeah, exactly.
And there is this really human side to solving crossword puzzles.
I loved the story of the woman who found her father's unfinished puzzles after he died.
Yeah, there's a really wonderful series in the Times called Solver Stories.
And there's one about the therapist, Deborah Stosen who, after her father dies, finds this portfolio of unfinished puzzles sunday puzzles that were all time sundays and she commits to finishing them and so she is seeing a kind of ghostly version of her now past father.
You know things he would know like classics.
All of those answers are filled in things he thought he knew but never really did, like car makes.
You know he's got wrong answers in and you have that weird moment of in pen going over her father's incorrect answers and kind of correcting them after the fact.
And then it's this fun collaboration across time, because he was the person who taught her about crosswords.
He schooled her in crossword ease, right?
He had a queen's accent.
So he would say like crossword ease, you know, things like epay and eek and etui.
These are the E words that appear in puzzles so often.
And it's this really lovely moment that that you know talks about puzzles as something that we don't just do together in the present, but that is an intergenerational process, form of transmission.
You know the knowledge that is passed along, the stuff that our parents know, that they teach us.
Yeah.
All of that is kind of held aloft in her, in her ritual.
Yeah.
You know, the thing with unfinished puzzles, I have so many unfinished puzzles.
I can almost never get every word.
How do, how do people who make puzzles feel about it?
If you're just going to Google and looking things up to complete the puzzle.
Is that cheating or is that like I'm going and learning things?
Yeah, I definitely think it's the latter.
I think that it's your puzzle and, you know, do with it what you will.
You know, I think that I write puzzles now on a monthly cadence for The New Yorker.
And those puzzles are hard.
They're very referential and elusive.
They talk about.
You know poets, that I like new literature, and not everyone's going to know the answers.
But the hope is that the puzzle ends up being a kind of like mixtape or recommendation engine or after the fact you didn't know a thing, but you look it up and you think I might, you know, get this book.
This sounds like an interesting person.
So I think it's definitely up to you what you do with the puzzle.
And I also think a lovely thing about crosswords is that it's a language unto itself.
And so I've seen people just get so much better at solving in a short span of time because they end up sort of cracking the code right.
Yeah.
And there's that code, the crossword ease and learning what all those words are and everything.
Well, I just loved the book.
Natan last, it's called Across the Universe.
If you're a Grammarpalusian, we're going to talk in the bonus section about the early history of crosswords.
Ben Zimmer told us some about it before earlier this year, but Natan has like so much more in his book.
It's just fun and fascinating.
So we're going to talk about that in the bonus section.
But Natan, thank you so much for being here.
Besides across the universe, where can people find you online?
Yeah, this has been lovely.
You can find me on my website, natanlast.com and all of the social places, X, Instagram and so on.
I'm natanlast.
Perfect.
Thank you so much.
