Google Code-in

Choose one of the languages Apertium uses in language pairs and prepare training data for the labeller from its UD-treebank: replace UD tags with Apertium tags, parse the treebank, create fastText embeddings. Then train a new model on this data and evaluate an accuracy.