Speedmantle

Guess the secret word: Speed Round

Every day, you're given word 1/1000. You then have minutes to find the secret word. The countdown begins as soon as you close these rules.

You can win by finding the secret word as quickly as possible, but you probably won't find the secret word! In that case, your final score is 1000-word ranking of your best guess. Good luck!

FAQ

Yes, .

Midnight UTC

I can think of at least four reasons for this.

Your guess, or the target word, is polysemous, and the meaning that is similar is rarely used. This is why "leather" is far from "patent." Sometimes one usage is simply more popular (among newspaper reporters, which is the corpus): "display" is more often a verb than a noun, and its vector reflects this.
You capitalized your word. SmartKey and some other keyboards stupidly ignore the autocapitalize settings that I have explicitly set in the HTML, and there does not seem to be anything I can do about this. I added a checkbox to help you avoid this.
Your word and the target word belong to different parts of speech. Sometimes this matters a lot. Sometimes it matters only a little.
By "similarity", we really mean "used in similar contexts". The principle was articulated by John Rupert Firth, who wrote, "[Y]ou shall know a word by the company it keeps." So, "love" and "hate" may seem like opposites, but they will often score similarly. The actual opposite of "love" is probably something like "Arizona Diamondbacks", or "carburetor".

The data set is what it is -- it's not perfect, and I can't afford enough computing power (or a big enough corpus) to try to make a better one. The technique has limitations. Sometimes, they'll bite you and you'll lose.

No old games, and no random games. There are three reasons for this:

Mindless imitation of that other word guessing game.
A game designer I know found a site that let you play as much of the other game as you want. He immediately binged, and then got bored, and hasn't tweeted about it since.
Another friend of mine said that she considered giving up on one day's Elmantle, and then realized it would be another 24 hours before she could try again. So she persevered and got it.

But you can send your friend a custom game. And maybe your friend will send you one back. Maybe the word will be flirtatious. Maybe you will conduct an entire romance over Elmantle. I met my wife through Zendo, so it's not impossible. Just remember the word limit.

Here's an article on word2vec, the underlying technology.

GoogleNews-vectors-negative300.bin from late 2021

Unusual word found! This word is not in the list of "normal" words that we use for the top-1000 list, but it is still similar! (Is it maybe capitalized?)

No clue why the data set doesn't include them. Perhaps they were just warping the scores of everything else, since they are used in so many different contexts?

The data seems to be normalized to US spelling. Elmantle tries to automatically Americanize your spelling (in the cases where only the American version is in the data set). Of course, they probably couldn't normalize some words, like 'biscuit', 'lift', or 'pants', so for those you're on your own.

Yes: click on the "#" column header. Click it again to go reverse-chronological. It will revert to similarity order when you enter a guess (or when you click the "Similarity" column header).

I was caught between a rock and a hard place here: on one hand, ranks are usually descending, but on the other hand, here, the similarities are ascending. I chose what I thought was the least bad option, but I don't think I would have been happy either way.

I grabbed a random list of the "most popular" 5,000 words in English, and removed anything capitalized or with hyphens, and the word2vec stopwords ("and", "if"). Then I shuffled it.

I'm still trying to figure out the right design, but probably eventually.

Dark mode

Treat all input as lowercase (see #2 from the answer to this FAQ)