Disease named entity recognition from biomedical literature using a novel convolutional neural network

To put this into perspective a great comparison of different word-CNN/LSTM char-CNN/LSTM combinations with and w/o CRF layer is given in this paper https://t.co/RtYMNEOAKe Yet, they always predict the tag, not the tag + prev/next tag.

Efficient and SOTA NER in clinical NLP: no bi-LSTM-CRF, just local predictions for prev/curr/next and (apparently) decoding only valid combinations. Are there analogous results in a more generic domain? https://t.co/JXrGc8t1ix