Guess the Secret Word!
March 15, 2022 12:31 AM   Subscribe

Semantle has been mentioned a few times in MeFi Wordle-type-game threads—it works by letting you know how "semantically similar" the word you've guessed is to the target word: how close you are, or "(cold)". From the maker, erstwhile MeFite David Turner: "...[I]t's not about the spelling; it's about the meaning. The similarity value comes from Word2vec. The highest possible similarity is 100 (indicating that the words are identical and you have won). By "semantically similar", I mean, roughly "used in the context of similar words, in a database of news articles." He also warns that you will need more than six guesses. You can play once a day, but you can keep on guessing during that game's 24-hour period. It'll keep some stats on how you've been doing over time, too.

From the date of this posting:
Q: Can I see yesterday's word?

A: Yes, "track".

Q: What words were closest to yesterday's word?

A: tracks, racetrack, levigated, cuppy, racing, record, speedway, field, pace, race, in descending order of closensess [sic]. More?

Q: Source code?

A: Yes!
I freakishly hit the target word in ten tries, once. Usually it takes me over eighty.
posted by not_on_display (76 comments total) 18 users marked this as a favorite
 
I hit the target word in 24 tries yesterday and don't think I'll ever manage in fewer tries than that.
posted by bigendian at 12:41 AM on March 15, 2022


Got to within three words pretty quickly, then gave up when I realized I didn’t have any intuition for the distances and wanted to stop randomly picking words. “The data set is what it is”.
posted by migurski at 12:58 AM on March 15, 2022 [1 favorite]


The dimensionality is too damn high.
posted by away for regrooving at 1:23 AM on March 15, 2022 [5 favorites]


I got one of the words recently in 11 tries but while I'm getting better that was not indicative of skill. I got a couple of lucky guesses.

What's more interesting to me is how to approach the middle game when you may have numerous green words but you're not getting anywhere with the theme you've established. At that point it seems like a good idea to go back to semi-random guessing to try and hit upon another connotation of the target which you may have missed.
posted by solarion at 1:26 AM on March 15, 2022 [3 favorites]


This is kind of wild.
posted by nubs at 1:27 AM on March 15, 2022


Pimantle is Semantle with a fancy visualization. It helps with distinguishing multiple "themes", making the game a bit less obscure.
posted by reynaert at 1:56 AM on March 15, 2022 [17 favorites]


solarion: agreed, regarding the middle game. Yeah.

Like the board game Concept, this feels like a good game to cooperatively play with people whose mental models of the world are pretty different from your own.
posted by brainwane at 2:19 AM on March 15, 2022 [4 favorites]


46 for today’s. I stumbled close after a lot of flailing because I tried out a dumb joke.
Looking at the list of 1,000 closest words didn’t give me much insight into how to approach this game.
posted by Mister Moofoo at 2:55 AM on March 15, 2022


LMAO it took me 105 tries.
posted by cendawanita at 3:39 AM on March 15, 2022


Thanks reynaert, I much prefer Pimantle's UI!

On Semantle, I got within 470 words on the third try but then I kept missing the context and it took me forever to get closer. After having played Pimantle (different word!) once I got close enough (within 20 words) then the answer was basically a synonym. But that took me like 60 tries!
posted by vacapinta at 3:49 AM on March 15, 2022


Each time I’ve tried this I’ve got well into three digits with no sign of progress, and the couple of times I’ve got it were a complete fluke.

It’s throwing random darts at an invisible target as far as I can tell, at least how my brain works.
posted by grahamparks at 3:55 AM on March 15, 2022 [1 favorite]


fuuuuuuck 57. it's awful I can't wait to inflict it on my boyfriend
posted by 5_13_23_42_69_666 at 5:17 AM on March 15, 2022 [2 favorites]


I got it in 31 for today's word. I thought that was horrific (because I had like 8 guesses over 500/1000) but apparently it's not too bad? I'll have to play some more to see how bad at this I really am, though.
posted by which_chick at 5:24 AM on March 15, 2022


I'm doing pimantle and edging closer and closer slowly, but honestly I can't figure out the links between "lift" and "limp" being quite close, and then "damaged" being closer still (though still only #289).

I'm assuming the swirly branches have some kind of meaning (Limp and lift are on separate branches, which converge at damaged)
posted by Just this guy, y'know at 5:29 AM on March 15, 2022 [1 favorite]


I've been playing this since I saw it first mentioned here a month or so ago. There are some secret words that seem easier to get than others because they are more closely related to nearby words than other secret words. Common words that have many different meanings seem the hardest so far.

Of course, what words you guess in what order plays a huge role in how long it takes. My first green guess for today's word was at 961, which sounds great, but that set me on a merry chase down a long false path. My wife still hasn't gotten it, and she's got way more green words than I did. We're both over 200 guesses. In contrast, we both got yesterdays in 15 guesses.
posted by mollweide at 5:31 AM on March 15, 2022 [2 favorites]


I got today’s in 25, but that required some leaps - and then when I looked at the “closest words” list it became clear that whatever data set this is pulling from, it associates the very common and not at all specifically sports-related target word with a particular sport. I don’t think that’s a spoiler because I would never in a million years have made that association.

Basically: this is not looking for neutrally DEFINITION-based similarities, rather it seems to be looking for words that frequently appear together or in similar contexts.
posted by showbiz_liz at 5:41 AM on March 15, 2022 [1 favorite]


After 16 guesses, I have 997/1000 and after 8 more guesses, I have no further idea.
posted by jacquilynne at 5:48 AM on March 15, 2022 [1 favorite]


On guess 20 I had 981/1000, I'm now nearly at 100 guesses and no closer. Totally bewildering.
posted by knapah at 6:01 AM on March 15, 2022


82 for today's, but yesterday I did it in I think 26 and was unreasonably proud.
posted by wellred at 6:33 AM on March 15, 2022


I was introduced to this game a few days ago and while I'm very good at most word games, I can't manage to wrap my brain around this one. I'm just chucking words out there and it's hit or miss and the feedback isn't enough for me?
posted by kimberussell at 6:37 AM on March 15, 2022 [1 favorite]


I did get the word once. Generally I go till I get a green box. Basically it's way too hard and life is too short. Most of the time I can't even see the connection between the target and what I've got.
posted by bifurcated at 6:57 AM on March 15, 2022


It was weird. I got it in 36. But there was a huge jump between my 6 guesses in the top 1000 and the final answer. I went from a similarity of 26.48 to 100. The other high scoring words I had clustered around what I thought was a different concept altogether. I just tried something I thought was different but maybe had some weak conceptual link.
posted by Schmucko at 7:04 AM on March 15, 2022


The word2vec thing is really important for doing well at this game. Over and over your brain will try synonyms, which can kind of help, but what you really want to do is try other words that can replace the green word in a sentence in the same kind of context.

Anyway it still routinely takes me more than 100 guesses.
posted by congen at 7:07 AM on March 15, 2022


This is maddeningly addictive. I’m on 35 guesses and have gotten within 253 words; then my next guess will be back to single digit similarity. I have a theme going, I think, but am not making progress right now. May have to try again later.
posted by TedW at 7:42 AM on March 15, 2022


POW! 37 guesses! Somehow stumbled on it by chance, it seems like.
posted by TedW at 7:45 AM on March 15, 2022


Got to 997/1000, but finding the connections extremely unintuitive.
posted by knapah at 7:52 AM on March 15, 2022


So, I finally started typing in random words that were largely unrelated to 997 and finally hit on the correct word which is essentially antithetical to 997. I get that it's not meant to find synonyms, but I clearly don't understand how it works.
posted by jacquilynne at 8:04 AM on March 15, 2022 [1 favorite]


So, I finally started typing in random words that were largely unrelated to 997 and finally hit on the correct word which is essentially antithetical to 997.

Well… sort of? Based on the connected word list, 997 isn’t a synonym but it’s also the same TYPE of word, if you see what I mean. To use an unrelated example, ‘hot’ and ‘cold’ are opposites but they’re a lot more closely related than, say, ‘tiger’ or ‘effervescent.’
posted by showbiz_liz at 8:13 AM on March 15, 2022


I've been able to figure it out most days, and I can usually see the connection between my near guesses and the correct answer. But after I've solved it, when I look at the list of nearest words, I can sometimes sense weird biases in the body of Google news articles that generated the connections.

In particular, I think the source corpus was heavily weighted towards sports. As showbiz_liz notes, in today's Semantle, a sports word that doesn't seem related to me has somehow made it into the 990s (IE, the ten most closely related words) Meanwhile, science words that I think are intimately related are all the way down in the 550s.

That's not a criticism-- it's just an illustration that any machine learning is going to be biased by the thing you trained it on. And keeping the bias in mind while I make my guesses is part of the fun.
posted by yankeefog at 8:20 AM on March 15, 2022 [7 favorites]


I got "surgeon" a few weeks ago in under twenty guesses and was like "welp I think I won semantle" and closed the tab. Haven't been tempted to reopen it since. That thing is hard.
posted by potrzebie at 8:32 AM on March 15, 2022 [1 favorite]


83
posted by It's Raining Florence Henderson at 8:50 AM on March 15, 2022


But then I tried it on my phone so I'd have it bookmarked for later and somehow got it in 1.
posted by It's Raining Florence Henderson at 8:56 AM on March 15, 2022 [1 favorite]


Based on the connected word list, 997 isn’t a synonym but it’s also the same TYPE of word, if you see what I mean

Early on I tried a word that is very much the same type of word as the answer (I got there eventually) and it got rated cold with a similarity of around 16. Turns out that the plural of that word is #801. And is not too different to the word at #981 to me.

Anyway, the dataset is the dataset, but all I'm saying is arghhhhh.
posted by knapah at 8:58 AM on March 15, 2022 [1 favorite]


Hold on a second--

Track:
tracks, racetrack, levigated, cuppy, racing, record, speedway, field, pace, race

Okay, I see the first two answers, sure, but "levigated?" And then cuppy? Why do How should By which way WHAT
posted by JHarris at 9:12 AM on March 15, 2022 [2 favorites]


I was already cuppy as hell after the third Bloody Mary so I levigated a couple of NoDoz to keep me upright before heading to the track.
posted by It's Raining Florence Henderson at 9:34 AM on March 15, 2022 [4 favorites]


Got today's in 28, and based on the comments here I think I should never play it again.

If I could remember how to do the "click to reveal hidden text" thing here, I'd do that, because I really don't understand how the words that were close (462/1000, 753/1000, and 815/1000) were related and I would like y'all to explain.
posted by Spathe Cadet at 9:58 AM on March 15, 2022


I love semantle. It's way more fun if you're not really concerned with how many guesses it takes; I usually clock in around 50-90 guesses. Today I got the secret word in 69, which was nice.
posted by phooky at 10:07 AM on March 15, 2022 [2 favorites]


Oh, this is David Turner's other game: can you name 10 words that have as little to do with each other as possible?
posted by not_on_display at 10:19 AM on March 15, 2022 [4 favorites]


Interesting, but I think I’ll wait until I’m sentenced to a long spell in prison, in which case it will be a godsend.
posted by Turtles all the way down at 10:41 AM on March 15, 2022 [1 favorite]


To use an unrelated example, ‘hot’ and ‘cold’ are opposites but they’re a lot more closely related than, say, ‘tiger’ or ‘effervescent.’

Right, but you would still expect warm to show up as more related to hot than frozen and that doesn't seem to be the case as often as not.

I think from looking at the related word list, it is clear that the problem is that the training data set really likes hockey.
posted by jacquilynne at 10:42 AM on March 15, 2022 [2 favorites]


997 isn’t a synonym but it’s also the same TYPE of word, if you see what I mean

a sports word that doesn't seem related to me has somehow made it into the 990s


The first guess I had today that was close was 270 which sent me down a long path that was crazy cold. Looking through the complete list of nearby words I start to see how it came to be associated via 365 and 837 and 683; though the keyword I would expect to link all of those together is not on the near list at all. 997, for me, is associated with the sport that would bring it all together; in short, I can see certain news articles covering the sport using all these (270, 365, 837, 683, 997), other sports related terms in the 990s (and elsewhere) and todays word.

Knowing what this used as a training set is really important, I think, because that helps narrow down the possibilities and guesses; the system behind this has been "trained" to see certain associations that may seem common to it based on the data set it has. In short, this is a fun game that is more about understanding the possibilities & limits of machine learning than anything else, at least from where I am sitting - its a word association game in which we try to understand how an alien, non-native English speaker who has done a mathematical analysis of a series of news articles, perceives English to work.
posted by nubs at 11:13 AM on March 15, 2022 [2 favorites]


First timer. Took me 51 tries to crack the top 1000. (Ice was 'cold'- ha!)
91 tries to get in the top 500, but got it in 97.
Interested to try again tomorrow.
posted by MtDewd at 11:20 AM on March 15, 2022 [1 favorite]


Track:
tracks, racetrack, levigated, cuppy


From Miriam Webster definition: cuppy: full of small depressions - a cuppy racetrack (the sports article bias this seems to have)

Levigated - guessing it has to do with the possibility of the term meaning something being smoothed, polished - i.e., perhaps it was used in the opposite way to cuppy to describe a track?
posted by nubs at 11:25 AM on March 15, 2022 [1 favorite]


I just got today's word in 19 guesses.

Based on past experience with Semantle, this is a feat I will never, ever replicate.
posted by box at 12:13 PM on March 15, 2022 [2 favorites]


Yea, I hate it, but came back to the tab several times to keep guessing.

Agree that training data set was NOT medical omg!
posted by esoteric things at 12:18 PM on March 15, 2022


Yeah, this is a complete no for me. The first time I tried it, I randomly ended up in the high 900s right away, and got it in about 50 guesses. Ever since then I haven't even come close. It's too frustrating, so I'm giving it up.
posted by BlahLaLa at 12:39 PM on March 15, 2022


Got it in 43, very brain bendy. Will be attempting tomorrow as well.
posted by Mizu at 1:07 PM on March 15, 2022


Got it in 20 the first time I tried. Better never try again.
Pimantle is worse for me... I don't understand that either, but it feels as if I should, and that makes me angry. I ragequit that one.
posted by Too-Ticky at 1:45 PM on March 15, 2022


The new Semantle is live!
posted by mollweide at 5:12 PM on March 15, 2022 [1 favorite]


@not_on_display, no, that's not mine -- I do link to it as prior art, but I only remembered it when a friend mentioned that she was enjoying playing "inverse" Semantle and trying to find the furthest word.

Glad everyone is having fun. I had to optimize some and upgrade the server last night to handle the midnight UTC spike, so hopefully everyone's experience is better today. It's been quite a ride, accidentally creating a game that (as of last check) 185,000 people were playing every day.
posted by novalis_dt at 5:33 PM on March 15, 2022 [15 favorites]


WHAT
I found the second one I tried in 8 guesses! 😳
posted by Too-Ticky at 1:25 AM on March 16, 2022 [2 favorites]


holy crap I got today's in 16
posted by 5_13_23_42_69_666 at 1:49 AM on March 16, 2022 [1 favorite]


Still struggling with today's, but learned that misspellings are accepted if common enough; "cemetary" placed in the top 1000, although "cemetery" was a lot closer.
posted by TedW at 5:57 AM on March 16, 2022


Well over a hundred guesses and still failed yesterday's. Then I win today in six!

I feel like it should only count tries subsequent to hitting a 750+/1000 guess. Or just start you off with a couple of 500s.

Would be fun if you could point this at a big set of documents, or even a website to build your own set of words. Be fascinating to see if it's a lot easier if set up on something you're very familiar with.
posted by Jobst at 7:05 AM on March 16, 2022


Dang, I got today in 17 and was very proud til I saw 6 and 8. This game is great.
posted by Measured Out my Life in Coffeespoons at 9:50 AM on March 16, 2022


I feel the same way as a pal of mine on Twitter "I hate this game, I play it every day." There's something about being able to get better at something that involves kind of... unfocusing your eyes to figure out how the words you DO know are related to one another. That day when the word was "pork" and I had hit a high one with "seafood" then kept hitting listing nearly every kind of seafood there is figuring eventually I'd land on it? Took forever. I think my average guesses so far is 120 (that was a stat I suggested) and I freaked out the other day when Daylight Savings in the US now means it starts an hour later. Seven was a good time for me! Eight is less good!

Anyhow, I am glad this dumb game exists because I kind of need a timewaster in addition to making a few unloved neurons fire.
posted by jessamyn at 9:54 AM on March 16, 2022 [5 favorites]


It's amazing how much I can get distracted on a garden path of words. I gave up today after 180+ guesses only to find I was so close but just couldn't get out of the tight rut of similarities I'd excavated. I was close to jumping out of it, though, I only found out after revealing the answer.
posted by not_on_display at 10:40 AM on March 16, 2022 [1 favorite]


Give UP? This I do not understand. Why would you give up when the internet is so richly stocked with ways to cheat? There's the thesaurus, there's putting all your top scorers into search and mining the results for other words, there's the OED. And then there's the Semantle reddit thread with all the hints.
posted by Don Pepino at 11:53 AM on March 16, 2022


Wait, is that a hint?
posted by not_on_display at 2:09 PM on March 16, 2022


So I tried this and thought it was OK, maybe harder than a Wordle variant since it took a few guesses more but whatever, and then I started reading here and realized I was incredibly lucky in my initial game. Now I'm really curious to see how tomorrow goes.

I solved Semantle #46 in 8 guesses. My first guess had a similarity of 21.90. My first word in the top 1000 was at guess #8. My penultimate guess had a similarity of 6.01. https://semantle.novalis.org/
posted by Cris E at 2:15 PM on March 16, 2022


I solved Semantle #19 in 8 guesses, mostly due to incredibly luckily getting 997/1000 on my second guess. More usually, I get 200 guesses in and give up half the time, and get it in about 150 the rest of the time.
posted by NMcCoy at 9:19 PM on March 16, 2022


115 guesses for #46, and I'm glad it doesn't track elapsed time.
posted by ManyLeggedCreature at 9:00 AM on March 17, 2022


I did #47 in 71 guesses. I figure it's a win any day my average drops. Maaaaaybe I am getting the hang of this!
posted by jessamyn at 10:00 AM on March 17, 2022 [1 favorite]


After a frustrating run of far too many guesses the past few days, I got today's Semantle in 29. Surely this means I'm getting better! HAHAHAhahalollolsniffsob.
posted by mollweide at 6:48 PM on March 17, 2022


I got #48 in 97 guesses and I'm pretty pleased with myself that I got it at all. Not bad, for a non-native speaker.
posted by Too-Ticky at 8:03 AM on March 18, 2022 [2 favorites]


I'm close but not close enough. Hmmmm not synonyms...shit.
posted by Oyéah at 2:01 PM on March 18, 2022


You found it in 86! The secret word is ______.
😑
This is the one I should have found in ten.
posted by not_on_display at 12:49 AM on March 19, 2022 [1 favorite]


I finally got one! Took me 23 guesses. (I tend to give up after 100 or so, so I run out of patience long before the game runs out of words.) I was hovering around one particular sense of the word, imagine my surprise when I looked at the list of closest words and they were all about a completely different meaning of the word. My nearest guess was 928/1000 before I got it.
posted by Daily Alice at 10:29 AM on March 19, 2022 [2 favorites]


112 guesses today, but I cheated by using a thesaurus. By the end, I had guessed 998, 997, 994, 992, and 990, and I was running out of synonyms.

There's an entry in the FAQ about hints, and I wonder what would make a good hint system for a game like this. Maybe if you could look at the first letter after enough guesses or after a close enough guess -- sometimes you have a pretty good idea of the answer, but you just can't hit on the right word.
posted by ectabo at 6:21 PM on March 29, 2022


I maintain it's not cheating the use the thesaurus (mainly because it only sometimes helps). The subreddit often has people who are offering hints hidden behind spoiler tags and I've found them useful occasionally.
posted by jessamyn at 6:28 PM on March 29, 2022


I went 350 words the other day without cracking the top 1000, and 650 words before giving up. I think my vocabulary is too large sometimes and my random word association attempts to break into top 1000 territory would be less fraught if I wasn’t trying words like inveigle, gloam, Carolingian and peripatetic. After playing maybe 15 games, I’ve finally won more games than I’ve given up, but I’m usually around 250 words+ when I do solve.

I really have to get more general and the blind alleys of similes just kill me sometimes. This game frustrates book smart and rewards inductive reasoning. Yet I count down to the new game assiduously every evening.
posted by Devils Rancher at 3:26 PM on April 2, 2022


I just can't with this game. Maybe if it started you out with a randomly selected word out of the top 1000. The neat thing about Wordle is how even if you start out with a word with none of the right letters, that rules out almost 1/5 of the possible letters, and a couple of wrong guesses give you enough information to deduce the right answer most of the time. With this one you have to be wiling to put in a lot of time guessing random words until you get one that's even in the right neighborhood. Ain't nobody got time for that.
posted by Daily Alice at 5:32 PM on April 2, 2022 [2 favorites]


Today I had no guesses in the top 1000 until randomly getting the word in 35.
Yesterday I had 40 of the top 100 but did not get it.
Every day is different.
posted by MtDewd at 7:20 PM on April 3, 2022


I have deduced, after winning today’s game at 455 guesses (I had been slowly circling in since my first hit at 150) that my sole asset with this game is an implacable, dogged determination. I question even my basic literacy at this point, but I’ll be back tomorrow at 7 pm CDST.
posted by Devils Rancher at 8:26 PM on April 3, 2022 [1 favorite]


There is now a "make your own" option. As novlis_dt said in the Reddit thread, it's "wildly untested" but kind of fun.
posted by jessamyn at 5:48 PM on April 8, 2022 [2 favorites]


Only now seeing this is by Mefi's own, yay! We've been enjoying this as a slow-moving thing to work on at intervals and chitchat about during the day.
I keep thinking, what I want is a hint function for when I get tired of guessing words outside the top 1000, that will give me a randomly chosen word somewhere at the bottom of the top 1000. (Maybe a random pick from among #1000-#950.)
posted by LobsterMitten at 7:06 PM on April 11, 2022


« Older Best golf movie ever? Caddyshack. Worst golf movie...   |   Big Boat Stuck Again Newer »


This thread has been archived and is closed to new comments