Combines multi-task learning and instance selection to reduce biomedical NLP training data while preserving competitive performance.
Transfers biomedical domain and task knowledge from English to lower-resource languages through cross-lingual sentence-embedding distillation.