Some Tildes users already know about this game from this thread. It was quite early on when I shared it there and I thought it deserved its own post. The game is called Noun Sense. It's a game...
Some Tildes users already know about this game from this thread. It was quite early on when I shared it there and I thought it deserved its own post.
The game is called Noun Sense. It's a game about word association. Each day, you get 10 adjectives. Your job is to guess the most common noun that follows them.
I created the dataset on my own laptop off a large collection of text files (billions of words). I generated the adjective/noun pairings and their frequencies utilizing a variety of sources (magazines, news articles from the last few years, movie subtitles etc). I have a full blog post about it if you are interested in the development details.
Lately, I've been busy tweaking the UI and handling user migrations from guest -> regular account. I also added a friend system, streak system and leaderboard.
Give it a shot here and let me know what you think!
If college was free I would have definitely taken more anthropology. It's so interesting how much culture shapes vocabulary. I'm abysmal at this because the word sources aren't internet culture in...
If college was free I would have definitely taken more anthropology. It's so interesting how much culture shapes vocabulary. I'm abysmal at this because the word sources aren't internet culture in the slightest. Feudal wasn't 100 japan, and korean wasn't 100 bbq and that just doesn't compute for me lmao.
Feels like a single player version of Family Feud, and I can see the appeal to it. It’s not my personal cuppa, but it’s good: it looks nice. It plays well. You can tell what’s going on at all...
Feels like a single player version of Family Feud, and I can see the appeal to it. It’s not my personal cuppa, but it’s good: it looks nice. It plays well. You can tell what’s going on at all times. It doesn’t take long to pick up. Well done.
Thanks! I am actually working on the coop multiplayer version right now. The plan is to have a full coop mode, where you work together as a team to guess the top 10 nouns in a given period of time...
Thanks! I am actually working on the coop multiplayer version right now. The plan is to have a full coop mode, where you work together as a team to guess the top 10 nouns in a given period of time (or maybe cumulative score?). I'll probably also create a free for all version or team vs team (family feud style!). I appreciate you checking it out.
Looking at the top collocations for each word, I am a bit confused by some. For example, why is it "multiple time" and not "multiple times?" And what is an "aforementioned nigh?" Googling it gets...
Looking at the top collocations for each word, I am a bit confused by some. For example, why is it "multiple time" and not "multiple times?" And what is an "aforementioned nigh?" Googling it gets zero results.
Yeah, that's more of a display issue. Multiple time and times count the same. I use word stemming for equality but then the "top answers" can look a bit awkward. Previously I did not show the top...
Yeah, that's more of a display issue. Multiple time and times count the same. I use word stemming for equality but then the "top answers" can look a bit awkward. Previously I did not show the top answers so this wasn't an issue.
I was also quite confused by the "aforementioned nigh" today.
Seems it leaned a bit too heavy into movie subtitles for this one and misunderstood it as a word... Will need to do some searching in the text dataset as to how that ended up as the top answer.
100%, I think it was some weirdness in the subtitle dataset... Where an acronym was standing as a noun and then punctuation was stripped. From the quote: "Therefore, it is also the five triangles...
100%, I think it was some weirdness in the subtitle dataset... Where an acronym was standing as a noun and then punctuation was stripped.
From the quote:
"Therefore, it is also the five triangles composing the aforementioned N-I-G-H, each are equal to the triangle C-A-B in this geometric concept!""
I have a daily standup at work where we often play word games as a group to try and get our brains working in the morning. Things like NYT Connections, Bracket City, Costcodle, etc. So this is...
I have a daily standup at work where we often play word games as a group to try and get our brains working in the morning. Things like NYT Connections, Bracket City, Costcodle, etc. So this is right up our alley.
We played today's version and really liked it, we scored about 700/1000 which we were pretty proud of! The feedback from everyone was that they would enjoy adding it to our rotation of morning games. Nice job!
I’m rank 2 for today (ok the puzzle only released 20 minutes ago) It’s fun. I did see that US vs UK spelling could have the same answer appear twice. And also: the share button should include a...
I’m rank 2 for today (ok the puzzle only released 20 minutes ago)
It’s fun. I did see that US vs UK spelling could have the same answer appear twice.
And also: the share button should include a link and not just a photo. Other games always have the answers in text, allowing people you shared it to (me and a friend play other daily games often) to play together.
Just an idea.
Fun game!
Fun! It reminds me of Semantle. Sometimes I get the secret word in 30, sometimes it takes hundreds. It sent me down a whole Word2vec rabbit hole, and I even tried amateurishly incorporating...
How to play?
The objective is to guess the secret word.
Each guess must be a single word. Semantle will inform you how semantically similar your guess is to the secret word.
Unlike other word games, this game is not about spelling; it's about meaning. We calculate this meaning using artificial intelligence (specifically word2vec technology).
In word2vec, each word has a measurable semantic distance from another, indicating their level of relatedness. Once you get within one thousand words of the secret word, we will tell you in the proximity column.
Sometimes I get the secret word in 30, sometimes it takes hundreds.
It sent me down a whole Word2vec rabbit hole, and I even tried amateurishly incorporating Word2vec into my work as a language query writer for a customer service department.
Not sure if this has already been mentioned, but on my iPhone (iOS 26.3) with Safari and the default iPhone keyboard, when typing in the words, if I use predictive text and select a word, nothing...
Not sure if this has already been mentioned, but on my iPhone (iOS 26.3) with Safari and the default iPhone keyboard, when typing in the words, if I use predictive text and select a word, nothing seems to happen: the selected word doesn’t show up in the input field.
The more I play, the more I like this. But the more I would also like to have a clear idea of what the dataset is. Who I’m trying to emulate.
iPhone has been a bit notorious for having some issues with auto-complete. I messed around with it again this morning with mixed results. Doesn't seem to have any issue at all on Android; I'll...
iPhone has been a bit notorious for having some issues with auto-complete. I messed around with it again this morning with mixed results. Doesn't seem to have any issue at all on Android; I'll keep trying to fix it but no promises as I've already spent some time on it.
As for having a clear idea of what the dataset is... ideally you wouldn't need to try to emulate any specific "persona". It should just "work" and feel fair. I have a bit of dirty data related to subtitles I still need to clean up... but then hopefully it will feel a bit better just giving your first gut reaction and scoring well (assuming you are a native speaker).
In case it helps, I did some testing and it seems that the issue is related to the capitalisation of words. If I start typing a word on the iPhone keyboard with capital letters (e.g. NAT) and then...
iPhone has been a bit notorious for having some issues with auto-complete. I messed around with it again this morning with mixed results. Doesn't seem to have any issue at all on Android; I'll keep trying to fix it but no promises as I've already spent some time on it.
In case it helps, I did some testing and it seems that the issue is related to the capitalisation of words. If I start typing a word on the iPhone keyboard with capital letters (e.g. NAT) and then select a suggested all-capital word (e.g. NATIONAL), this gets added into the input field without issues. However, if I do the same with lowercase letters (e.g. type "nat" and then select "national"), the word doesn't get added.
This is REALLY helpful. This was actually the crux of the whole issue. I have a fix for it coming that instructs the iphone keyboard to use capitalization when doing the auto-correct. I tested it...
This is REALLY helpful. This was actually the crux of the whole issue. I have a fix for it coming that instructs the iphone keyboard to use capitalization when doing the auto-correct. I tested it with an old iphone 7 and the auto-correct worked for me every time. I really appreciate you finding this. I have a similar issue in a few of my other games so this really helps a lot!
I'll try to push the fix for it later this evening along with the data cleanup from the subtitles issue.
Neat. The problem I encountered was that after creating an account, I had to play again and I forgot some of the words and got a significantly worse score :( but at least I got 100 on the first...
Neat. The problem I encountered was that after creating an account, I had to play again and I forgot some of the words and got a significantly worse score :( but at least I got 100 on the first entry and another one.
Yes. When I followed the account verification link, I got a new puzzle. When I went back to the other window, it updated with the new puzzle (new as in post-verification, not a new day).
Yes. When I followed the account verification link, I got a new puzzle. When I went back to the other window, it updated with the new puzzle (new as in post-verification, not a new day).
I really like this! Is there no way to reset a forgotten password, though? I registered an account but the password didn’t get saved in my password manager so now I’m stuck 🙁
I really like this! Is there no way to reset a forgotten password, though? I registered an account but the password didn’t get saved in my password manager so now I’m stuck 🙁
Hey, I have not implemented that on the daily site yet. That said, you should be able to use my main site's forgot password here I still need to get that implemented in the daily game styling so...
Hey, I have not implemented that on the daily site yet. That said, you should be able to use my main site's forgot password here
I still need to get that implemented in the daily game styling so it isn't so jarring with the style clash. Eventually I will move everything over to this new codebase on the daily site as I like the aesthetic much better.
When doing the forgot password, It will take you through the flow with my main site's styling... the username/pwd is shared between the two. Logging in on the main site will also keep you logged in on the daily one. Apologies for the confusion! A lot of stuff in transition at the moment, I was hoping someone wouldn't run into this yet.
Agreed, that would be the best case scenario. I originally had typo forgiveness in place... so if you typed 'tcao' instead of 'taco' it would fix it for you and allow it. (was using something like...
Agreed, that would be the best case scenario. I originally had typo forgiveness in place... so if you typed 'tcao' instead of 'taco' it would fix it for you and allow it. (was using something like the Levenshtein distance) I guess in most situations, that might also work for American english to British english (like behavior -> behaviour or color -> colour)
Note: I'm not trying to imply American vs British spelling is a "typo". Was just thinking of a simple solution.
I may see if I can put that back in a smarter way... the original implementation had some issues where it could potentially match on a totally different word as it saw it as a "typo". e.g. face -> ace. (1 letter removal typo).
I think if it ends up with a score of 0 after checking, then maybe I'll allow typo check... with some additional algorithm that would be smarter and not reward a totally different word.
Out of curiosity, does the dataset take spelling variations into account, or would "colour" and "color" be considered different words for something like BRIGHT x? In addition to "aforementioned...
Out of curiosity, does the dataset take spelling variations into account, or would "colour" and "color" be considered different words for something like BRIGHT x?
In addition to "aforementioned nigh" that someone mentioned here from today's answers, I was wondering about "bent copper", which seems to be the best answer for "bent". It's interesting that it ranks highest, considering that it is (I believe) so strongly a British term. As a player, I might want to know more about the sources from which these phrases come. My guesses would be different if I knew that the sources were mainly films subtitles, as opposed to if they were mainly online news sources. I was also wondering if the English Wikipedia might be the most neutral source and the easiest for players to calibrate their answers for?
I wonder if something like this could actually be a useful language learning (vocabulary building) tool. Maybe not for beginners, but for something like B1 or B2 level learners.
I really like the game. It took me far too long to realise why its name is what it is!
Really great suggestions and interesting insights. I used a variety of sources for the data. It was a combination of news articles (I think this was the heaviest weight), movie subtitles, and...
Really great suggestions and interesting insights. I used a variety of sources for the data. It was a combination of news articles (I think this was the heaviest weight), movie subtitles, and wikipedia (and a few other things) so it should have quite some variety of different versions of English.
The answers that bubble up to the top can be a bit surprising to me as well. I think for infrequently used words like 'aforementioned' which is quite a formal there is some difficulty. My movie subtitle dataset in particular may have had some duplicates (same movie, different file names). I didn't deduplicate this at the time. So when the word is less frequently used, but used in a movie, it can unexpectedly become the "top answer" due to low frequency usage in other datasets. I'll try to clean up the dataset in the coming week or two and hopefully it will fix these issues.
For your question about British English, after some thought, I think it's actually interesting if those are separated. I may implement the fallback if a score would be zero... (if color was found, but not colour, you can still get some points). Leaving them separated allows the actual "common" phrase to bubble to the top. The aim is for it to be fair for everyone.
I think I will also create a "top community" answers area that can show for previous days. It would be interesting to see what the consensus for each adjective is from real people. (Maybe it will become part of the dataset if it wasn't already included?).
The top community answers could be really interesting to see! Maybe players could even earn extra points if their answers aligned with other players in past days.
The top community answers could be really interesting to see! Maybe players could even earn extra points if their answers aligned with other players in past days.
The Google books n-gram data might also be an interesting source data set for this, especially since it distinguishes word classes, like copper_ADJ and copper_NOUN. A downside if that it is a...
The Google books n-gram data might also be an interesting source data set for this, especially since it distinguishes word classes, like copper_ADJ and copper_NOUN. A downside if that it is a rather excessive amount of data and therefore inconvenient to work with.
Some Tildes users already know about this game from this thread. It was quite early on when I shared it there and I thought it deserved its own post.
The game is called Noun Sense. It's a game about word association. Each day, you get 10 adjectives. Your job is to guess the most common noun that follows them.
I created the dataset on my own laptop off a large collection of text files (billions of words). I generated the adjective/noun pairings and their frequencies utilizing a variety of sources (magazines, news articles from the last few years, movie subtitles etc). I have a full blog post about it if you are interested in the development details.
Lately, I've been busy tweaking the UI and handling user migrations from guest -> regular account. I also added a friend system, streak system and leaderboard.
Give it a shot here and let me know what you think!
If college was free I would have definitely taken more anthropology. It's so interesting how much culture shapes vocabulary. I'm abysmal at this because the word sources aren't internet culture in the slightest. Feudal wasn't 100 japan, and korean wasn't 100 bbq and that just doesn't compute for me lmao.
Very fun :D
Special thanks to @balooga for some great feedback and bug finding. And thanks to early adopters @cutmetal and @HiddenTig.
Feels like a single player version of Family Feud, and I can see the appeal to it. It’s not my personal cuppa, but it’s good: it looks nice. It plays well. You can tell what’s going on at all times. It doesn’t take long to pick up. Well done.
Thanks! I am actually working on the coop multiplayer version right now. The plan is to have a full coop mode, where you work together as a team to guess the top 10 nouns in a given period of time (or maybe cumulative score?). I'll probably also create a free for all version or team vs team (family feud style!). I appreciate you checking it out.
Looking at the top collocations for each word, I am a bit confused by some. For example, why is it "multiple time" and not "multiple times?" And what is an "aforementioned nigh?" Googling it gets zero results.
Yeah, that's more of a display issue. Multiple time and times count the same. I use word stemming for equality but then the "top answers" can look a bit awkward. Previously I did not show the top answers so this wasn't an issue.
I was also quite confused by the "aforementioned nigh" today.
https://www.google.com/search?q=%22aforementioned+nigh%22
Apparently it's from the movie "Better Off Dead"
https://youtu.be/8PwHypXvsEU?t=38&is=8fmdtPS_HVkr68Q1
Seems it leaned a bit too heavy into movie subtitles for this one and misunderstood it as a word... Will need to do some searching in the text dataset as to how that ended up as the top answer.
It must be "aforementioned night" surely??
Surprisingly not.... That also appears as a separate answer in the database. I'll get it cleaned up soon.
FWIW nigh isn’t a noun, so that isn’t adjective-noun pair.
100%, I think it was some weirdness in the subtitle dataset... Where an acronym was standing as a noun and then punctuation was stripped.
From the quote:
"Therefore, it is also the five triangles composing the aforementioned N-I-G-H, each are equal to the triangle C-A-B in this geometric concept!""
Noun Sense and enclose.horse from that linked Tildes thread have bumped NYT Spelling Bee down to my tertiary daily game.
I have a daily standup at work where we often play word games as a group to try and get our brains working in the morning. Things like NYT Connections, Bracket City, Costcodle, etc. So this is right up our alley.
We played today's version and really liked it, we scored about 700/1000 which we were pretty proud of! The feedback from everyone was that they would enjoy adding it to our rotation of morning games. Nice job!
Glad you all enjoyed it. Thanks for trying it out!
Friend code sharing thread! Here’s mine. The leaderboard’s more fun when you have some rivals to compare yourself against 😉
Great idea!
Here's my friend code NHT7NV. (you'll need a regular account and to be logged in to click the link)
Here's mine!
Y6G2BB
I’m rank 2 for today (ok the puzzle only released 20 minutes ago)
It’s fun. I did see that US vs UK spelling could have the same answer appear twice.
And also: the share button should include a link and not just a photo. Other games always have the answers in text, allowing people you shared it to (me and a friend play other daily games often) to play together.
Just an idea.
Fun game!
Fun! It reminds me of Semantle.
Sometimes I get the secret word in 30, sometimes it takes hundreds.
It sent me down a whole Word2vec rabbit hole, and I even tried amateurishly incorporating Word2vec into my work as a language query writer for a customer service department.
I tried this one the other day and was terrible at it. It felt less intuitive and "fun". Love the idea, but somehow I didn't enjoy it all that much.
Not sure if this has already been mentioned, but on my iPhone (iOS 26.3) with Safari and the default iPhone keyboard, when typing in the words, if I use predictive text and select a word, nothing seems to happen: the selected word doesn’t show up in the input field.
The more I play, the more I like this. But the more I would also like to have a clear idea of what the dataset is. Who I’m trying to emulate.
Thanks again for the game!
iPhone has been a bit notorious for having some issues with auto-complete. I messed around with it again this morning with mixed results. Doesn't seem to have any issue at all on Android; I'll keep trying to fix it but no promises as I've already spent some time on it.
As for having a clear idea of what the dataset is... ideally you wouldn't need to try to emulate any specific "persona". It should just "work" and feel fair. I have a bit of dirty data related to subtitles I still need to clean up... but then hopefully it will feel a bit better just giving your first gut reaction and scoring well (assuming you are a native speaker).
Glad you are enjoying the game!
In case it helps, I did some testing and it seems that the issue is related to the capitalisation of words. If I start typing a word on the iPhone keyboard with capital letters (e.g. NAT) and then select a suggested all-capital word (e.g. NATIONAL), this gets added into the input field without issues. However, if I do the same with lowercase letters (e.g. type "nat" and then select "national"), the word doesn't get added.
This is REALLY helpful. This was actually the crux of the whole issue. I have a fix for it coming that instructs the iphone keyboard to use capitalization when doing the auto-correct. I tested it with an old iphone 7 and the auto-correct worked for me every time. I really appreciate you finding this. I have a similar issue in a few of my other games so this really helps a lot!
I'll try to push the fix for it later this evening along with the data cleanup from the subtitles issue.
Neat. The problem I encountered was that after creating an account, I had to play again and I forgot some of the words and got a significantly worse score :( but at least I got 100 on the first entry and another one.
Interesting, it should have been migrated if you used the same browser. How did you register? Username/password by chance?
Yes. When I followed the account verification link, I got a new puzzle. When I went back to the other window, it updated with the new puzzle (new as in post-verification, not a new day).
I really like this! Is there no way to reset a forgotten password, though? I registered an account but the password didn’t get saved in my password manager so now I’m stuck 🙁
Hey, I have not implemented that on the daily site yet. That said, you should be able to use my main site's forgot password here
I still need to get that implemented in the daily game styling so it isn't so jarring with the style clash. Eventually I will move everything over to this new codebase on the daily site as I like the aesthetic much better.
When doing the forgot password, It will take you through the flow with my main site's styling... the username/pwd is shared between the two. Logging in on the main site will also keep you logged in on the daily one. Apologies for the confusion! A lot of stuff in transition at the moment, I was hoping someone wouldn't run into this yet.
Hurrah, thanks very much!
I think it should accept British English too
Agreed, that would be the best case scenario. I originally had typo forgiveness in place... so if you typed 'tcao' instead of 'taco' it would fix it for you and allow it. (was using something like the Levenshtein distance) I guess in most situations, that might also work for American english to British english (like behavior -> behaviour or color -> colour)
Note: I'm not trying to imply American vs British spelling is a "typo". Was just thinking of a simple solution.
I may see if I can put that back in a smarter way... the original implementation had some issues where it could potentially match on a totally different word as it saw it as a "typo". e.g. face -> ace. (1 letter removal typo).
I think if it ends up with a score of 0 after checking, then maybe I'll allow typo check... with some additional algorithm that would be smarter and not reward a totally different word.
Thanks for the suggestion!
Out of curiosity, does the dataset take spelling variations into account, or would "colour" and "color" be considered different words for something like BRIGHT x?
In addition to "aforementioned nigh" that someone mentioned here from today's answers, I was wondering about "bent copper", which seems to be the best answer for "bent". It's interesting that it ranks highest, considering that it is (I believe) so strongly a British term. As a player, I might want to know more about the sources from which these phrases come. My guesses would be different if I knew that the sources were mainly films subtitles, as opposed to if they were mainly online news sources. I was also wondering if the English Wikipedia might be the most neutral source and the easiest for players to calibrate their answers for?
I wonder if something like this could actually be a useful language learning (vocabulary building) tool. Maybe not for beginners, but for something like B1 or B2 level learners.
I really like the game. It took me far too long to realise why its name is what it is!
Really great suggestions and interesting insights. I used a variety of sources for the data. It was a combination of news articles (I think this was the heaviest weight), movie subtitles, and wikipedia (and a few other things) so it should have quite some variety of different versions of English.
The answers that bubble up to the top can be a bit surprising to me as well. I think for infrequently used words like 'aforementioned' which is quite a formal there is some difficulty. My movie subtitle dataset in particular may have had some duplicates (same movie, different file names). I didn't deduplicate this at the time. So when the word is less frequently used, but used in a movie, it can unexpectedly become the "top answer" due to low frequency usage in other datasets. I'll try to clean up the dataset in the coming week or two and hopefully it will fix these issues.
For your question about British English, after some thought, I think it's actually interesting if those are separated. I may implement the fallback if a score would be zero... (if color was found, but not colour, you can still get some points). Leaving them separated allows the actual "common" phrase to bubble to the top. The aim is for it to be fair for everyone.
I think I will also create a "top community" answers area that can show for previous days. It would be interesting to see what the consensus for each adjective is from real people. (Maybe it will become part of the dataset if it wasn't already included?).
The top community answers could be really interesting to see! Maybe players could even earn extra points if their answers aligned with other players in past days.
The Google books n-gram data might also be an interesting source data set for this, especially since it distinguishes word classes, like
copper_ADJandcopper_NOUN. A downside if that it is a rather excessive amount of data and therefore inconvenient to work with.It's possible that "bent copper" is the result of a misinterpretation of phrases like "bent copper wire".
Super fun!
799 :)
I'm number one today 813/1000 baby RAAAAAAH
I'm average! 725.