monsoon-nlp/dv-labse

This is an experiment in cross-lingual transfer learning, to insert Dhivehi word and word-piece tokens into Google's LaBSE model . The model currently outperforms dv-wave (ELECTRA) on the Maldivian News Classification task . Currently outperforms mBERT (Maldivian news classification) and dv MuRIL (a similar transfer learning model) The model is currently outperforming dv wave (Electra) and muRIL on the news-classification task, but may continue to train it further . The experiment is being conducted with no Thaana vocabulary and only 1,100 words from the model are added to the,