Inspiration
Machine translation is a topic I find very interesting because I've experienced the huge growth since the days when Google translate was of poor quality at best, to now, when translation quality can sometimes rival professional human translators. I also enjoyed reading many of Shakespeare's plays in high school and thought it would be interesting to combine the topics.
What it does
This model converts modern english sentences into Shakespeare-style sentences.
How we built it
This model was built using a transformer architecture, as described in the paper, Attention is all you need.
Challenges we ran into
One of the biggest challenges with this project was overcoming the small dataset size. The entirety of Shakespeare's plays consist of only around 50,000 sentences, which makes it difficult to train a transformer from scratch on. A couple things that helped included using sub-word tokenization, and increasing dropout during training to prevent overfitting.
Accomplishments that we're proud of
The translations are sometimes pretty effective! It's fun to play around with things like famous quotes, as those tend to produce the best outputs.
What we learned
It's pretty ambitious to take on a project like this without teammates! If I did it again, I'd get into a group so we could bounce ideas off each other and split the load.
What's next for YeOldeTranslator
If I had more time, I'd love to train it on a bigger architecture, use plays other than just from Shakespeare but also contemporaries, and try different more recent developments in transformer architectures.
Log in or sign up for Devpost to join the conversation.