A collaboration between Anna Garbier & Lan Zhang
"Toast" is a word guessing game, played between a computer program and one or more players. The game begins when the program chooses a mystery word from a large collection of words. The player tries to guess the mystery word by asking a series of "Is it closer to ___ or ___?" questions. For example, if the mystery word is schoolbus:
Player: Is the mystery word closer to toast or banana?
Program: Banana.
Player: Is it closer to banana or avocado?
Program: Banana.
Player: Is it closer to banana or the color yellow?
Program: The color yellow...
The guessing always starts with toast, hence the name of the game!
Behind the scenes, every word in program's "lexicon" is represented as a vector, and plotted in a two-dimensional space. In this space, words that are semantically similar to each other are clustered together; words that are semantically different are far apart. For example, "sofa" and "couch" are close together, but "sofa" and "orca whale" are far apart.
When the program responds to the player's questions, it does so by calculating the distance between the mystery word and the guessed words within the vector space. As the player moves through the space en route to the mystery word, a trail of guesses is drawn to the screen.
A trail of guesses, with the "hints" set to visible. A more challenging version can be played by turning off the visual hints.
The words were curated by us, processed with spaCy word2vec, scaled and plotted with Sci-kitLearn t-SNE, and are brought to your browser with pixi.js.
Word processing pipeline
Part of the goal of "Toast" is to explore linguistic datasets through play. The game currently has four datasets:
- Big bag of (random) nouns
- Animal names
- Alice in Wonderland
- Seinfeld
All four datasets, visualized
The food corner of the Seinfeld dataset
Creating a custom dataset
- Wordnet Morphy
- Mobile responsive