Add new resource for relation prediction (sebastianruder#435)

Update intent_detection_slot_filling.md (sebastianruder#421)

Add the result of Stack-Propagation.

add NER state-of-the-art result (sebastianruder#296)

Add NER state-of-the-art result with EMNLP 2018 paper (Learning Better Internal Structure of Words for Sequence Labeling), this paper proposed IntNet, which mainly focus on learning better character-to-word representations, IntNet significantly outperformed other character embedding models, and also combined with BiLSTM-CRF achieved state-of-the-art NER result without using any lexical features, transfer learning, language modeling or pre-training.