Too many times we've found that English is just too limiting a medium for communication. There are ideas that there are simply no words for. That's where we got the idea for Word Generator.
What it does
Generate novel and completely unseen, but totally believable words using an LSTM-RNN.
How we built it
We used an implementation of an LSTM-RNN in Torch and Lua trained on a massive 1.2 million line dictionary from gutenberg.org
Challenges we ran into
Training on a CPU was incredibly slow. But we eventually made progress. Also, preprocessing data was a big challenge.
Accomplishments that we're proud of
Creating really interesting words such as "Encyarorous," "Ravor," or "Rascterate."
What we learned
RNNs are really powerful!
What's next for Word Generator
Generate names for startups! Or generate names for fictional characters.