Bidirectional-Transformer-Machine-Translation

My final project for CS 479 Machine Translation at BYU. I chose to build a Transformer by programming each component using PyTorch and connecting them together to create the full architecture of an Encoder-Decoder based Transformer. The goal of this project was to become more familiar with the technical components of Transformers – since most of my experience with them had been abstracted to pre-trained models during my academic career – and build a bi-directional translator that could take English or Armenian input and translate into their Armenian or English, respectively. I built a pipeline to parse and clean TMX (Translation Memory eXchange) files from OPUS Parallel Data, Tools and Interfaces in OPUS (Tiedemann, LREC 2012) to extract English and Armenian sentence pairs to build a dataset as well.