Word Dance

A fast‑paced web game where you write sentences with a word missing so that an AI fills it in with a target word.

Web Dev
Game Dev
Truly one of the games of all time

Now, show me how you [MASK] with your words!

Live at

This is a web game in which the player must write sentences with a word missing, such that a machine learning model fills in the missing word with a target word.

Of course it’s a great match for English language learners, but you’d be surprised at how intellectually challenging it can be even for those who are fluent! I find that often the crazier the sentence, the more likely it is to succeed.


  • Programmed in Javascript using the React and Next.js frameworks
  • Currently deployed on Vercel

Do it yourself!

This game is INCREDIBLY simple to make! Check out the source here.

It could easily be the starter project in a Coding 101 level tutorial, since it’s just:

  • A single index.js file that handles all the game logic
  • An API route that calls the Hugging Face Inference API (can’t be called in the user-shipped JavaScript only because we don’t want to expose our authentication token)
  • And a couple of simple React components

If you’re sick of teaching beginners how to create React to-do list apps, I think Word Dance has a lot of opportunity for creativity and extension. There are a lot more interesting machine learning models on the Hugging Face Hub 😉

The Machine Learning Model

The machine learning model currently used is bert-base-uncased from the Hugging Face Hub

Authenticating with Hugging Face’s API

You will need your own HUGGINGFACE_API_TOKEN environment variable in order to authenticate with Hugging Face’s API and utilize their machine learning models.

You can add one in a .env.local file in the parent directory which Next.js will automatically load.