Semantle: Can you think like a computer?

Various xps: yeah i still don’t get how pretty tight synonyms can sometimes score so low - like, does the neural net write a different hierarchy of associations based on its selected primary definition? So if it’s looking for say, “croak” (frog sound) then it won’t be interested in “die” of “perish” - sorry if this is dumb or incoherent, still learning to think like a computer here

There is some mention on the site about polysemy* causing low scores - I guess the multiple uses mean they're found in different contexts more often and therefore aren't counted as semantically similar. Whereas a usual English-speaking brain would count the meanings additively, the computer penalises them via subtraction. Or something.

(Interestingly, 'polysemous' is not an accepted word in the corpus.)

Aw, nuts! I forgot to finish yesterday’s puzzle, which I woulda likely nailed in a few more guesses. My first guess was “love”, which I think was in the top 100? Green anyway. And I got to the 990s pretty quick. Anyhow, no braggin rights.

8 sleep -1.14 (cold)

are negatives supposed to be possible?

xps, I think it depends on what contexts it sees most in its corpus of example sentences, which the FAQ says are mostly taken from news articles?

So it doesn't deliberately pick a specific meaning of "croak" but it probably won't rate "croak"/"die" very high as it would be unlikely to appear with that meaning in many of its sample news articles. Meanwhile, there'll be a lot of words with more obviously similar meanings to "die" which it will have grouped "die" with instead.

Or the other day, when the word was "level" it didn't rate "flat" or "even" very highly although they're close synonyms in one not-very-newsworthy context but the 900s were full of words which might appear before "level" in news articles like "higher", "lower", "raise" etc.

(OK, the "before" part still mystifies me, regarding when it scores synonyms highly, vs not necessarily same-in-meaning but interchangeable-in-sentence words like antonyms, vs words which might appear nearby or just in the same sorts of article.)

This is basically what emil.y said except longer and less clear, but I don't think it deliberately subtracts scores for multiple meanings, it's just the algorithm doesn't understand multiple meanings at all, so the meanings it doesn't see as often don't rate highly at all. But I guess it may also penalise even the most common meaning, because there's effectively more noise in the data set from the meanings it doesn't know what to do with.

I solved Semantle #28 in 145 guesses. My first guess had a similarity of 11.67. My first word in the top 1000 was at guess #63. My penultimate guess had a similarity of 74.70 (999/1000). https://semantle.novalis.org/

I had guesses that had negative relations wtf

nice job...i struuuggled w/ #28, didnt even get in the green and tapped out @ 183 guesses

very stuck on 995

Re negatives: I adjusted the similarity algorithm today to be more correct (that's what I get for copy-pasting code!). And as of a few seconds ago I have adjusted the intro text to explain the range.

It's technically "cosine similarity" (multiplied by 100), if that means anything to you.

I solved Semantle #28 in 193 guesses. My first guess had a similarity of 1.94. My first word in the top 1000 was at guess #56. My penultimate guess had a similarity of 74.70 (999/1000).

this game makes me feel like a gigantic dumb lug googling "words related to politics"

Thanks for explanation David! And thanks for this awesome game!

158. Tough one!

tonight I had help from a friend who wanted to play, so it was more of a two-pronged effort.

I solved Semantle #28 in 647 guesses. My first guess had a similarity of -4.23. My first word in the top 1000 was at guess #315. My penultimate guess had a similarity of 16.84. https://semantle.novalis.org/

Took me over 300 guesses to get into the top 1000. with "animus" of all things

I solved Semantle #28 in 111 guesses. My first guess had a similarity of 2.99. My first word in the top 1000 was at guess #26. My penultimate guess had a similarity of 74.70 (999/1000)

fun one that, good journey

Jumped from not on the list to 999/1000, and then got the answer due to a habit of trying variations that has never paid off before.

gross today!!

I solved Semantle #28 in 109 guesses. My first guess had a similarity of 6.42. My first word in the top 1000 was at guess #6. My penultimate guess had a similarity of 74.70 (999/1000). https://semantle.novalis.org/

Fuck, I should have got it in the 20s, took me having political, philosophical, theological and cultural all up there before it clicked, and then it was so obvious. Gah. Not sure my insomnia-ridden brain helped either.

ok got it in an embarrassingly large no of guesses - smashed my head against a wall forever with half the top ten and totally stuck, then came back to it between sets at a gig and got it immediately argh!

After a lot of dithering in the cold mid-20s, I crashed straight into the top 1000 at 945, leapt immediately up to 999, and have been stuck there ever since.

I solved Semantle #28 in 98 guesses - my first successful one in a while.

My first guess had a similarity of 31.24 - thanks, I guess, to Putin-caused current events.

My first word in the top 1000 was ”hatred” at guess #13 - and I got to 425 at #29 with “propaganda”, but couldn’t get past there for ages. It was guess #81 ”polarize” that provided the necessary course correction to steer me into the home straight.

My penultimate guess had a similarity of 74.70 (999/1000). https://semantle.novalis.org/

leapt immediately up to 999, and have been stuck there ever since.

Fuck. Have got 993 and still can't crack it. This game is taking over my life.

OK, cracked it 115.

Well, that was a little lucky: I solved Semantle #28 in 22 guesses. My first guess had a similarity of 6.42. My first word in the top 1000 was at guess #9. My penultimate guess had a similarity of 53.67 (983/1000). https://semantle.novalis.org/

I got the 995 blues.

yeah me too.
Yesterday's I was so close to getting, I think I probably assumed I'd already done all versions of that word, but hadn't.

Oh, FINALLY. My big blind spot was discounting adjectives and assuming nouns - I must have guessed every "ism" there is.

I thought I'd guessed every ist or equivalent but I guess not.

the answer to this thread's title is: no, I can't. I give up.

I gave up. top guess was 'political', must have entered every political persuasion bar cornish independence; I did try to think more generally and I had 'dogmatic' but couldn't make the link.

I solved Semantle #29 in 107 guesses. My first guess had a similarity of 6.42. My first word in the top 1000 was at guess #82. My penultimate guess had a similarity of 47.24 (581/1000).

wacky one, took me 82 guesses to get in the green and then 4 guesses later i had 999, and then i honestly thought i might not get it. expected it to be something directly related to violins at least, especially after yesterday had 996-999 all some variation of the answer itself

Got it in 44. Pretty straightforward once you get into green

Gave up on #28 but got #29 in 44 as well. Guessed a cluster of words in the twenties having to do with writing, got from there to poetic, which was almost in the green, and from there it was an easy few jumps to lyrical --> musician --> music --> guitar --> piano.

Yep, followed a very similar route

I solved Semantle #29 in 21 guesses. My first guess had a similarity of 4.58. My first word in the top 1000 was at guess #9. My penultimate guess had a similarity of 8.79. https://semantle.novalis.org/

dumb luck but felt good

I always guess "tree" first. only cos it gets me going and often ranges between 8-10 percentile

holy fuck, that was the easiest one for me yet

I solved Semantle #29 in 39 guesses. My first guess had a similarity of 13.99. My first word in the top 1000 was at guess #39. My penultimate guess had a similarity of 0.71. https://semantle.novalis.org/

only got in the top 1000 once...when I guessed the word.

that is awesome!

all of my guesses had to do with vehicles, and then pedal almost got me into the top 1000, but other things like 'pedal' weren't working, so I thought of a pedal for a piano, and...that was it

I solved Pimantle #4 with 229 guesses and no hints!

https://semantle.pimanrul.es

someone made a clone with a 2D visualization, though i can't say i got any use out of it at all

I solved Semantle #29 in 48 guesses. My first guess had a similarity of 2.01. My first word in the top 1000 was at guess #12. My penultimate guess had a similarity of 77.56 (996/1000). https://semantle.novalis.org/

Phew, much less agony than yesterday. This game either provides a nice little lunchtime (AEST) diversion, or completely eviscerates my day.

After 400 guesses on #28 — and being stuck at 995/1000 — I went to a family get-together and couldn’t get back in time to SURELY solve that maddening thing. DNF! Then…

I solved Semantle #29 in 30 guesses. My first guess had a similarity of 6.01. My first word in the top 1000 was at guess #28 (923/1000 woot). My penultimate guess had a similarity of 54.14 (826/1000). https://semantle.novalis.org/

…which is the best I’ve done and it felt pretty good.

Neanderthal, mine was 28, 29, 30: accordion >>> keyboard >>> bingo!

I solved Semantle #29 in 80 guesses. My first guess had a similarity of 1.25. My first word in the top 1000 was at guess #40. My penultimate guess had a similarity of 58.69 (918/1000).

Fun one.

I wonder if the pimantle guy is a Mel Croucher fan.

I am now stuck on the 999/1000. I am absolutely baffled as to what it can be given that I guessed the words that obviously seem like the closest words.

I'm baffled that you're baffled! It's amazing how this game reveals the different types of semantic leaps people make. I didn't get 999 but I got a very similar one in the high 990s and then immediately guessed the answer. But then there have been games where I've had several in the high 900s and still given up.