Advances in Natural Language Processing (NLP) over the past decade have largely been driven by the scale of data and computation used to train large neural network-based models. However, these techniques are inapplicable to the vast majority of the world's languages, which lack the vast digitized text datasets available for English and a few other…