Neural negated entity recognition in Spanish electronic health records.

•The goal is to detect negated Clinical Named Entities.•Character embeddings are able to cope with lexical variability in EHRs.•The Bi-LSTM was proven robust to capture contextual information of the negation cue.•The system using lemma-embeddings outperformed its counterpart us-ing word-embeddings.•F-measure of 65.1 for exact match and of 82.4 for partial match in Spanish EHRs.