This research project aims to evaluate different techniques to translate between minority languages such as Irish and Scottish-Gaelic. The project’s focus is to use unsupervised learning i.e. translating with minimal to no parallel data. We implement and evaluate 3 techniques. 1) A model that creates a mapping between Irish and Scottish-Gaelic. 2) A model that combines word and character level knowledge using Deep Learning. 3) Facebook’s open-source model that learns a mapping using an adversarial technique. We observe that model 1 produced the best results with an accuracy of 50%. We apply an optimization to improve the accuracy to 85%.

The group would like to thank their faculty sponsor Dr. Kevin Scannell for their support of this project.