Hittite

One of our team members is taking a class about Hittite and wanted to make this difficult language more accessible to those who want to learn it. Thus, we created a translation tool for the Bronze-Age cuneiform language Hittite. We created a ML model that takes in images of Hittite characters and outputs its syllabic translation. Because there was not a dataset of Hittite characters and their translations, we created our own dataset of over 1500 data points - uploaded on Kaggle. Then, we created a GUI using Gradio that takes in both image uploads and user-drawn sketches and returns the syllabic translation. We hope this project will help increase interest in this ancient language and help preserve knowledge of it!

While developing this website, we ran into several challenges. We struggled with modifying the parameters of the model, integrating the sketchpad function on Gradio, coding with large amounts of data, and the lack of a pre-existing data set. Our initial idea was to have the users draw the character into the website itself. However, due to limited time, we were unable to accurately achieve this (approximately 50% accuracy rate). In the future, we hope to implement the drawing feature on our website. Though we went through many struggles, we are proud of the amount of new technology we tried to learn and use during this hackathon. This was all of our first times using machine learning. Overall, we would say this project was a very hard but rewarding process.

Built With

• gradio
• kaggle
• machine-learning
• python
• tensorflow