Anna Garbier

Toast2word V1 (Global)

Toast2word V1 (Global) is an AI remake of a much older word-guessing game. In this modern remake, a human and a computer program navigate semantic space together— one using intuitions, the other using vector math.

The program chooses a target word from a large list of nouns. The player tries to guess the target through questions and answers, beginning with “Is it more like TOAST or ___?” For example, “Is it more like TOAST or CAT?” If the target is more like CAT, then the question pivots: “Is it more like CAT or ___?” This procedure continues until the player arrives at the target.

When the program answers the player’s question, it does so by calculating the cosine similarity between word embeddings. The interface through which the player and program interact is a simple web form. The application is written in Python, using SpaCy embeddings, and Flask as a micro web framework. The code is on GitHub.

Toast2word V1 (Global) grew from an earlier collaboration with Lan Zhang. The newer V1 moves beyond constrained datasets by introducing a vocabulary of all common nouns, uses cosine similarity rather than euclidean distance, has a new interface, and is written entirely in Python.

More projects