Cross-lingual name entity recognition from clinical text using mixed language query

Abstract

Cross-lingual Named Entity Recognition (Cross-Lingual NER) addresses the challenge of NER with limited annotated data in low-resource languages by transferring knowledge from high-resource languages. Particularly, in the clinical domain, the lack of annotated corpora for Cross-Lingual NER hinders the development of cross-lingual clinical text named entity recognition. By leveraging the English clinical text corpus I2B2 2010 and the Chinese clinical text corpus CCKS2019, we construct a cross-lingual clinical text named entity recognition corpus (CLC-NER) via label alignment. Further, we propose a machine reading comprehension framework for Cross-Lingual NER using mixed language queries to enhance model transfer capabilities. We conduct comprehensive experiments on the CLC-NER corpus, and the results demonstrate the superiority of our approach over other systems.