Tibetan To Englsih Machine Translation

This project is the first step in the long-term project to create a Tibetan to English machine translation. The project aims to contribute to the preservation of Tibetan language and Tibetan Buddhism.

We first used SentencePiece to tokenize the data and then simultaneously ran three different approaches to train our data:

• We imported a transformer model from FairSeq, an open-source sequence modeling toolkit, and fed it with our own data and hyperparameters for training.

• We fine-tuned a pre-trained T5 transformer model provided by Huggingface Transformers library with our data.

• We built a transformer from scratch with Pytorch. Tutorial link: https://lionbridge.ai/articles/transformers-in-nlp-creating-a-translator-model-from-scratch/