Lexical simplification (LS) can decrease the communication gap between medical experts and laypeople by replacing medical terms with layperson counterparts. In this paper, we present: 1) a rule-based approach to LS using a consumer health vocabulary, and 2) an unsupervised approach using BERT to generate word candidates. Human evaluation shows that the unsupervised model performed better for simplicity and grammaticality, while the rule-based method was better at meaning preservation.
Keywords: Health Vocabulary; Lexical Simplification; Machine Learning.