Using Word2Vec to sort ungendered objects by gender.

Projections page 4

Using word embeddings trained on Twitter with the Word2Vec algorithm, I turned words into points in high dimensional space. I then calculated the vector from “boy” to “girl” and used it to sort collections of other words from most boyish to most girlish by flattening, or projecting them onto that vector.

I used this method to create a computer program which takes collections of previously ungendered words as input, and outputs graphs of the sorted words, where the amount of space between each one can be seen relative to the distance between “boy” and “girl.”

The result is a series of newly gendered things.

Read “Projections” here.