Slipping Into Semantics With Semantle’s Mastermind

Published Apr 30, 2024

Hi! Can you tell us who you are and about your game?

Good day to you all. I’m David Turner, also known as “Novalis”, and I’m the creator of the game Semantle—a unique word-guessing game that focuses on the meanings of words. Before I went into game making, I used to be a full-time programmer, and the software I’ve created and helped create is used daily by most people today. I still am a programmer by trade and my condition basically landed me on this path when I was a kid.

In Semantle, you don’t type out answers in the same way as you would in a normal word game. Here, players aim to find the secret word by making semantically related guesses. So the whole game is more on the meanings of words than the spelling of them. It can be very difficult, with some more avid players dubbing it the “Dark Souls” of word games, but it’s a challenge that is fun to pick at.

Can you tell us about your creative process?

So often the time when I do game design is actually as I'm falling asleep, it's a really good way to be thinking about something that is not a problem that I have to solve right now. It's not interpersonal conflict. And so either I will succeed in coming up with something or I won't. And it doesn't matter. I can fall asleep and I can try again tomorrow.

And so, Semantle, I was thinking as I fell asleep, okay, so there's Wordle. And then I played Globle, the one where you're guessing the country. And I thought, okay, so in that game, you're guessing and you get a direction. So you have some, like, two degrees of freedom there. And in Wordle, you have, I guess, five letters, and each of those has twenty-six ways you can go.

So what is something that has lots of degrees of freedom but not too many? And I remembered Word2vec, which has 300 dimensional vectors. Oh, that's great. This is a game, and either it's possible to play or it's not. There's just no way to know until you try it. But trying it is like nothing, right? It's just a couple of hours of programming. You download the thing and put together the most microscopic of friends. And I tried it and I found that it worked.

And so I sent it to three friends. I said, “I can't tell if I'm serious about this or not…”, and one of them posted it on his message board. And suddenly I looked up and I started having users. And in the little FAQ at the bottom, I linked to a friend of mine, and my friend wrote to me and says, “Hey, I'm getting a lot of traffic from your site.” So I looked at my logs.

Oh my god, there are 60,000 people playing this. And I sort of hadn't noticed because I put it up on the cheapest digital ocean server, and because it doesn't have anything doing, all it does is send you a vector. I was able to support a very large number of users on a very small server. So that was the process there.

Current project

For the thing I'm working on now, which is a graphical adventure game in the vein of Myst, the 1993 version with pre-rendered images. I've been actually collecting puzzles for this for over a decade.

And every time I have a puzzle, I write down ideas in my notes file, and sometimes I'll build a prototype and sometimes I won't. And sometimes the idea is vague. What about a maze that has a red part and a blue part? That's vague. But then to figure out how that goes together, sometimes you just have to program it and see what comes out.

Or I'll use Inkscape to draw things. Sometimes I'll draw things on paper if I absolutely have to, but I'd rather draw it in Inkscape because that gives me a way that I can rearrange things easily without having to think, I'm going to erase this and redraw it. I've been programming for so long that it's often the quickest way to get an idea somewhere. 

And on my board game designs, I think I haven't done that. I've often gone to a paper prototype with board games first because that's straightforward. But even there, sometimes there's a programming element involved.

As for the status of this graphical adventure game, it has not been launched. It's not even close. There's a lot of artwork that still needs to be done and some of the puzzles still need to be implemented. I have either prototypes of them or just have a document that describes how they work. But there's a lot of work, I think, probably a couple of years on that.

It's a different scope and it's a different style of game compared to Semantle, but it's a style of game that I absolutely love. And there aren't very many of them. I actually don't know why there aren't very many of them. Maybe the answer is that there's only a world market for a small number of them. But I think people who like them sort of like them enough that they're willing to play, even the ones that are kind of crappy.

And I think part of it probably is that it's hard to think about what is a puzzle that you can do that people will be able to solve without reading your mind. And it's not a puzzle that someone has already done before. So a lot of these, you play them as, and you wouldn’t be like, “Oh, it's a Klotski. That one where you're sliding the blocks around. I've seen that before.”

I'd rather see something that's a puzzle that I had never seen before that I really have to figure out.

You said you had a condition, what is it?

I have dysgraphia, so I have trouble with handwriting. It’s like, where my hand goes to write a “d” and a “g” comes out. It's not good. So my parents had me try to type things and got me a Thinkpad. 

And this is when Thinkpads were made by IBM, and they are absolute tanks. Very difficult to break, which, you know, for a kid is good. And they have on them what my friend likes to call the “Cursor Locator Interface Tool”. Some people call it the nipple mouse, the track point. And it's a little button between the g and the h keys that you use to move the mouse around. It's great because you don't take your fingers off the home row while you're typing.

You can just move it around. Not very good for precision, but very good when I just need to click a button while I'm in the middle of programming something. And my hands have become so molded to them that I tried using something else for, I think, two weeks. I instantly got carpal tunnel.

So now, whatever it is, I have to have a Thinkpad keyboard.

On preferred hardware

Well, it was IBM Thinkpads for me. Unfortunately, their supplier, Lenovo, bought them out. And actually, I just have to have the keyboard now. So I have a ThinkPad keyboard attached to a computer that some random person built and was trying to sell, so I bought it off of them. The keyboard is sold separately.

One of my first machines was the IBM 701 C. And if you get a chance to look that up, do so, because that's the one with the butterfly keyboard where it had a ten-inch screen but a twelve-inch keyboard. And the way they did that was when you closed the lid, the keyboard had split into two parts which sort of folded and rotated to fit under the screen. It was wild.

Can you tell us about the creation of Semantle?

It actually started quite close to its final form right now, except visually. The right thing to do at that time was that, like, I should have had it on a scale from 0 to 100 and scaled according to how close the closest words were. So the closest words should be at like 99 or something. But I did something different at the start and I wasn’t sure it would work.

But it turned out to work just fine. And my initial test was I’d pick a random word from the dictionary. There's some work that goes into the word list, but initially I think I was just using my local word list. So every Unix system, which is what I use, comes with a word list and it's 70,000 words.

It's not what you'd call a good word list, but it's a working word list. So I picked it up, and had a word everywhere. From there, I said, “Can I solve this? How many guesses is it going to take me to solve this? Can I even get close to solving it?”

And it turned out that I could. Once I had solved one, I said, “Okay, well, if I can solve one, then I can send that one to my friends, and then my friends see how my friends do at solving it.” And I sent it to three people who are good at puzzles, and one of them, I don't even think he tried it. I don't know. Another one did say that she had gotten to it.

And then my third one, he said, “I can't get this. I'm going to send it to my friends” And that was how it started. He put it on a message board with his friends.

On the difficulty of Semantle

While it seems a bit hard to understand, it is intended to be intuitive. You could imagine playing this with your friends in the car with one of you playing the role of the judge. Is this closer? Is that closer? And that was actually, I think one of the sort of background processes that went into this game. 

I had, around twenty years ago, heard about a game that worked that way, and the creator had called it French Toast. And I knew exactly who I had to tell about this, so I went ahead and told my friend Mark. He says, “Oh, yeah, me and Ranjit invented that ten years ago.” And, you know, Mark is brilliant. I absolutely believe that this is true. He says. He called it “Plenty Questions”.

But his version is to compare two different things. It's a little different. It's played with humans instead of being played with this absolutely impartial, perfect machine or maybe a completely arbitrary machine. In particular, the machine is trained on news articles. And so one of the cases where this comes up is a word like “belt”, where the only time belt comes up in a news article is in the context of boxing, where you have, like a championship belt. And so the sports-related meaning of belt, as opposed to a punch or a large drink or the roadway around Washington, DC, the sports one happened to be a little bit more dominant.

So, Semantle, I actually think, is probably on the hard end of games that people play of this sort. And I wasn't really thinking about how I could adjust the difficulty of this. The hint button is the way that ultimately I came up with to ease the difficulty. 

So it's sort of just luck that this ended up being playable at all. For other games, often the difficulty is adjustable. So one of the games that I made after Semantle is called Deco Deck. It's a logic puzzle where you're trying to group cards into groups of five. And there's only one way to do that and it expands the entire board.

And there, for the difficulty, it’s like, I'll set a computer to solve it and see how many steps the computer takes to solve it, how many blind alleys it goes down, how deep it has to go in its reasoning process. And there it's very easy to adjust the difficulty and I’ll be like, I'll have an easy puzzle, and that'll be a Monday puzzle, just like the New York Times crossword. And I'll have a hard puzzle, and that'll be a Friday puzzle. And then you can pick, you know, what level you do, and you can work your way up to the hard puzzles. And of course, the other way to adjust the difficulty of Deco Deck is the size of the puzzle.

A small puzzle is going to be easier than a large puzzle. But Semantle, there wasn't a good way to adjust the difficulty. It really was something that was almost more discovered than invented.

Do you have any design or game creation guide that you took inspiration from?

The sort of standard advice that I've heard on puzzles specifically is that you as a designer don't know how hard they're going to be. So in cryptography, there's a rule that anyone can create a cryptosystem that they themselves cannot break, but that doesn't mean someone else can't break it. 

In puzzle design, anyone can create a puzzle that they can solve because they know how their own brain works and they will make the right associations. So to create a puzzle that other people can solve, you really have to test it. Now, there are certain sort of narrow genres of puzzles where the paths have been worn smooth enough that if you make a crossword puzzle, you generally can just send it off to Will Shortz. And Will Shortz has looked at nine million crossword puzzles over the course of his career

One of the experiences I had that helped me out on this was in 2011. My team won the MIT mystery hunt, which is a collection of 100 to 150 puzzles that are solved over the course of a long weekend in large teams. And the prize for winning is a year of hard labor. You get to write the next year's hunt. And so I wrote only three puzzles for that hunt, but I edited about 50 of them.

And I test-solved a bunch as well. And there was one in particular I remember that I looked at, and what happened is the guy wrote about 20 puzzles and then said, “I'm going on vacation. You test them, you solve them. Tell me if they're solvable.” I looked at it, and I said, “No one is ever going to solve this puzzle.”

That puzzle was impossible. But he wrote it, and he's on vacation. I can't make any changes to it. I'll send it out to the test solvers, and the test solver did it. So I was very impressed by that. I thought, okay, so I genuinely don't know what a hard puzzle looks like. All I know is, this puzzle is hard for me, or this puzzle is easy for me. And so I try it, I send it out, and then I think, what do we need to change here? What hints do we need to give? 

And one puzzle that I was working on, test solving, there was a hint. There was a physical puzzle that had an electronic thing, and you soldered the thing together. And then there was a button and an LED, and you looked at the button, and you pressed the button, you looked at the LED, and you had to figure out what to do with that. And it turned out that the input was Morse code, and the output was Morse code. Morse code is not a difficult code, but it's a code that's difficult if you don't know it. You have to do it in real time as you're watching this LED blink and as you're pressing the button. And so I solved that half of the puzzle.

Okay, so what do I do now? And they gave me a hint. There's a component that's not soldered in. Aha! That's the hint I needed. And so I can't remember if we ended up giving that hint to other people during the hunt or not. But at least now we know, okay, this is the place where people get stuck on it. And therefore, if we need to make it easier, this is the place where we need to make it easier.

Could you tell us more about the other games you’ve created?

I’ve got something to share about Surf Words. I sort of wanted it to be hard because my model was Super Hexagon. If you've ever played Super Hexagon, it's a very simple game. You're a little triangle in the middle of the screen, and there's a hexagon moving in on you, and one of the sides of the hexagon is missing, and that's the side you have to aim for. And so your controls are rotate left or rotate right.

And the first time you play it, you're like, okay, I survived for five seconds, and the next time, three seconds, right, and then eventually, oh, I've got ten or fifteen seconds. And that's a good game of Super Hexagon. Now, if you're the person who created it, Terry Kavanagh, you can go on YouTube and you can find a video of Terry playing it, and he'll just sit there and play it for, I think, as long as he wants, because he's just that good at that kind of game. And in Surf Words,. I'm not quite at that level. But if I'm not talking to someone else while I'm playing it, if I'm not paying attention to someone else, I can survive for a minute. I feel good about surviving for a minute. And I also have it speed up a little as you go, like Tetris basically, because I want you to feel stressed out as you're playing this game. I don't want you to relax.

Where's the slowdown button? Actually, there are controls in there that let you set the speed. And if after three times, you've done, like, really poorly, it says, “Hey, you seem to be having a hard time, would you like to slow it down?” And in particular, the version where you only show two letters in advance is a lot easier than the version where you show three letters. Three letters, I think, is the true version of the game, because that's where you really have to put your word skills to use.

It's not that you have to know a lot of different words, because the game has several word lists and it always tries to keep you within the prevalent meanings of commonly known words. But if you do know lots of words, it's happy to follow you into the incredibly difficult words.

In fact, the computer, in some sense, has to be ahead of you at every stage, because when it's giving you a new letter, it has to give you a new letter that is compatible with what you have already done, if that is possible. And it tries to give you a letter that is compatible with a word that we believe that you are going to know. But if the only letter that is compatible only because you've gone off on some weird path, then we will give you a word that maybe you don't know because maybe your vocabulary is smaller than mine.

On improving vocabulary

I just read a lot. I do the New York Times crossword every day. And I actually started the crossword because I thought, well, maybe it will make me better at trivia because I'm a member of Learned League, the online trivia league. And at the time I was in the C level and the levels are A through E. And so C is officially mediocre.

And I wanted to get a little bit better. So I thought, well, maybe if I do the crossword, it'll sort of introduce me to who is that tennis player? And I think, like twice I've come up with an answer in Learned League that I can trace back to having done the crossword. So it absolutely did not help with that. The only thing the crossword is good at is helping me be good at the crossword.

And now I know who Arthur Ashe is. But I have somehow managed to get better at trivia. Well, I was in A, but then I lost. Got relegated this last round, so I'll be in B next time. Some of that's practice and some of that's just reading more.

Could you tell us about your monetization scheme, if there were any?

I didn't put ads on it. I don't really care for ads. I understand that that's what there is now after I sold it, but, like, I run an ad blocker. I recommend that you run an adblocker as well. All of your listeners should be running an adblocker, and I like uBlock Origin

You know, monetizing was not necessarily in my list back then. I had, at the time, a full time job , and my expenses were very low. I think towards the end I had to put it on a bigger machine, and I might have been paying $50 or $60 a month for it, but $50 or $60 a month was not going to break the bank.

So I could keep running it like that forever. Monetizing was not really on the list. But then I started getting a couple of offers, and the offers were large enough that I said, okay, I would like to do this. Maybe I'll sell this. And then maybe if I sell this, I could make more games.

And so I did.

Peak traffic

A quarter of a million unique IPs every day was the peak. And then I think unfortunately it's gone down since then. I wish it were more popular, but I totally understand why people don't want to fry their brains every single day before work. And also, of course, at the time I released it, it was sort of towards the tail end of the lockdown portion of the pandemic when people were at home a lot. And now I hope that people are going out and seeing theater instead of playing games or if they are playing games, playing games with their friends.

That was one of the things that surprised me about Semantle was that I got a lot of emails from people saying, I am playing this with my friends, I am playing this with my family. And when I thought about it, I thought, that actually made a lot of sense. You end up getting a lot more divergent ideas. What is a word that is really different from all the words that we have tried so far?

And it's more fun to be stuck in a group, in part because you have everyone else egging you on and you can't give up because everyone else is still there trying to solve it.

On the acquisition of Semantle

I actually had two offers; one of them was sort of vague, and the other one was extremely concrete. And I said, “Here's how much I would like to sell it for.” And he said, “Well, I can't quite make that, but I would offer you another amount.” And it was an amount that was going to, let's call it a year of salary, or something like that.

Now I can quit my job, right? And obviously, I don't have to quit my job. I could keep my job and keep making games on the side. But after talking to my wife, I decided that I would try and make a living at this and see how it goes. And one of the big advantages of this is now if my kid has a day off school and I have to hang out and stay home with her, that just means my game comes out a day later and my wife can continue to work and make the big bucks.

After the acquisition, the guy said to me, “Listen, I want to keep you. Make sure that you're available in case I have questions about how to do things.” And he did write to me with a few questions. Especially early on, you had, “How do I do this?” “How do I do that?” 

And then recently he had a question and I set up the necessary documentation and explained how to do things. I'm available because I want the game to succeed and I continue to have sort of a little bit of profit sharing from it. But obviously, I want the game to succeed regardless. It's got my name on it after all!

What does it take to be successful in the game creation process in your opinion?

So I guess one thing I would say is that there's something that's simple, but it's not easy, which is to do something that no one else has done before. A lot of people like to start by taking a thing that someone has done and building off of that. And that's how you get from the base 2048 to all of the various 2048-like games. Some which are actually kind of interesting, you know? There’s this one that’s it’s 2048, but it's real-time. Then there’s another 2048, but it’s sort of crossed with Tetris. 

And that's certainly a way that you can go. But if you can go away off into the forest where no one else has gone, then you can often sort of pretty quickly figure out, does this work? Does this not work? And if it does work, then you have something to consider, which is how to sell it to people.

For example, your elevator pitch would be like, “Well, it's a cross between Super Hexagon and a war game" which I recognize is starting from something else, but it's a new kind of logic puzzle. That's your elevator pitch. And it's a hard pitch.

“It's a game you've never seen before!” “It's so hard it'll make your eyeballs bleed!” stuff like that. But also, getting to testing pretty quickly is good. You know, your first prototype is you test with yourself. If you're building a board game, you sit down and you are two players or three players at the game and making it function on your own.

Your second test, after you are satisfied with how you can play with yourself, is with other human beings sitting down in a room with them. And this can be a very interactive test. You can say, “Okay, well, this isn't working.” Midway through the game, I'm just going to cut it off, or this isn't working, I'm going to change that rule.

A friend of mine brought me a game that his friend had created, and it sort of had two sub-games. You were collecting cards and then you were using them to fight. The game’s objective was supposed to be getting the best collection compared to the other players, but the actually fun thing turned out to be the card battle mechanic. And midway through the test, he says, “Okay, so this game is originally not supposed to be about the fighting.”

So why don't we just agree not to fight and see how the other half of the game plays, just to see how it works? And I don't know what happened to that game. Maybe the answer is that they decided to make it all fighting because that was the fun part. Or maybe the answer is they decided to eliminate the fighting altogether. They're in a sort of a tough spot, because if the fighting is the fun part, well, you can’t take that part out or it’d be boring.

What do you think is the fine line between inspired homages and dull copies?

It’s hard to tell the line between, oh, this is a remix of an old thing and, oh, this is a completely new idea that has never before been seen in the world. I think, in general, taking a thing you love and trying to make it better is a valid way of doing things. Or taking a thing that you wanted to love but actually just couldn't. And trying to build a good version of it can often be a productive strategy. There was a talk I gave where I heard someone give a talk and I said, “I really wanted to like this talk, but I didn't.”

It didn't go anywhere. I'm going to give the version of this talk. There's the version that I wanted to have listened to.

Would there be any changes to Semantle if you were given the chance to restart it all over again?

Yeah, I think there would be a few things. I think one of the big ones would be the similarity score. You actually don't really need to give a similarity score to two decimals on this weird scale. I think I should have just given a score from zero to 100 scaled according to how similar, the most similar words are today. The most similar words were, I think at like 60 or 70. But sometimes the most similar words are at like 30 or 40. So I think that's a change I would make.

I think I would focus a little bit more on the visual design early. The early versions were very bare bones, very web 1.0. I'm not primarily a visual person, so if I can, I'd like to work with a visual designer. And I think that would have made a difference. I didn't expect to release it. Like I said, I sent it to three friends. That's how it started. So if I had expected to release it then I would have put the design in first.

And these days when I'm testing something I say, “Hey friends, I'm sending it to you, but I'm not releasing it yet.” And when it’s all green then I say, “Okay, fine, it's time to release.” Now that I have at least something that resembles a visual design, that's great advice.

Are you fond of any game in particular lately?

My favorite game is a word game called Montage. It takes exactly four players in partnerships and the way it's sort of a cross between a Taboo and a crossword puzzle. So you have a grid, like a crossword puzzle, and you have, some of the spaces on the grid that have colored dots, each of which represents a five or six-letter subset of the alphabet. 

So the orange one is ABCD and Z. And imagine you have a three-letter word, and the first letter is orange. So now you're going to give your partner a clue, and your clue has to be five or fewer words, so I might say household pet. And again, your first letter is going to be ABCD or Z. So I'm thinking “cat”, but my partner says “dog”, but they both match so either of those would be acceptable. Now, my partner has to guess this by knocking on the table before both of the opponents knock on the table. If both of them knock, then I have to pick one of them to guess.

And you're doing this all within a 1-minute timer, so it's very stressful, which I really enjoy.

As for games, back when I was around twelve or thirteen, Myst came out. That was something that absolutely grabbed me. I played some Doom and some other games, but then I discovered the best game of all was you could program your own computer, and it had the hardest puzzles because the puzzle was trying to figure out how to make this thing do the thing you want it to do. And I sort of never looked back. I have clinically bad handwriting, so I had a laptop in school when I was a kid, and, you know, in theory, I would take notes on it and practice programming on it.

So I learned something, but not necessarily the thing that they were trying to teach me. And then I found, sort of fell into a job, a programming job, and have enjoyed doing that ever since.

Other hobbies

I do enjoy ceramics. Although, since I've been making games more, the ceramics have fallen a little bit by the wayside. But my brother-in-law just dropped off a keyboard and says he wants some sort of ceramic enclosure for it. So we'll see how that goes.

What’s next for you?

Yeah. So a few months ago, I came out with middlesgame.com. The way it works is that I give you the middle of the word, and you have to figure out what's on the front of the word and what's on the back of the word to complete the word. Usually, I try and choose something that has a middle that is unique among commonly known words. So the example I give is “-ob-” is in the middle of only one word that you probably know, which is “bobcat”.

It's also in the middle of mobcap, which is some old-timey kind of hat that no one's heard of. I hadn't heard of it until I crept through my word list. So we keep on letting you guess. Okay, what's at the beginning of this and what's at the end of this until you can figure out what the whole word is. And the game gives you ten guesses.

And because there's only one commonly known word, your score is really just the number that you have. And you can do it in one guess if you're good at figuring out what the middle of a word is. But most people will try a few things at the beginning and the end until they figure out what the word is going to look like.

So I've tried a few things to get traction on various games, and frankly, none of them have worked very well. You know, I had a publicist send out a press release on Deco Deck. I'm really proud of it. And I thought, “Oh, this is going to be one that's got mass appeal. It's really pretty. Everyone will love it and it just absolutely sank without a trace.” 

I think I tried buying some Google Ads and that did nothing. It's interesting. Google makes a lot of money from ads so someone must be buying them and therefore someone must be clicking on them, but I haven't found it.

I wish I had advice on a foolproof way to drive traffic to your game. I know that people do TikTok and that kind of stuff. I did post on Reddit, Twitter and some others. Or maybe I just need the right people to find it, or maybe I just need the right game. And I think with the graphical adventure game, there's a community of people who are into these sorts of things and people who are into them will find it.

So I think it has a better built-in audience. And so that’s the hope there. But it's also a tremendous amount of work because you have to have all the graphics that put you into a place and not to mention all the testing. You're testing a dozen or two dozen puzzles instead of testing one puzzle.

I’m hoping to release it by the end of 2026, but that’s it. Hoping to. We'll have to see how the art comes along because that's really the thing that is, I think, the long pole there.

Where can we go to learn more about Novalis and his games?

If you go to Novalis, you’ll find me and you'll find all of my games. And there's an email address there and you can find me on activity pub. That's Mastodon for most people. And I love getting emails. I've gotten some wonderful emails about Semantle over the years.

People, old friends, people who say that they play it with their family. I love getting emails, so keep them coming!

Have a game to sell?

Let’s find out if we play well together.