Homolinguistic Translator

Many years ago I came up with the idea for a "homolinguistic" translator after learning about the Stanford NLP GloVe project at a conference. The project uses machine learning to create vector representations of words from a data set, which is a fancy way of saying that the words become hypothetical points that can be used to calculate mathematical distance. I wanted to use the tool to transpose input text into a version of itself that was a slight evolution of the original message by changing randomly selected words. To put it more poetically: I wanted to play telephone with my computer.

I implemented my idea almost immediately but I kept it to myself, running it on my computer as a creative tool for getting myself out of linguistic ruts. The idea was to, through the act of machine approximation of meaning, manufacture the feeling of my own language becoming alien to me. By making my words "other," I could take a found language approach to my own work and enact a kind of recycling process on my own ideas.

I've had a number of requests to share this program online and I've finally figured out the best way to do it. The program is purposely slippery—there is no guarantee it will respect the spacing of your text, and it will give you a different version of the translation every time you run it. Just type your text into the input area and click "translate."