NCRFpp

Why we included this project

NCRF++ is a PyTorch toolkit for sequence labeling tasks like named entity recognition, part-of-speech tagging, and word segmentation. The appeal is that you never write a training loop: the network is assembled from a configuration file, pairing a character-level encoder (LSTM, GRU, or CNN) with a word-level encoder, and using softmax or CRF for inference. You can also mix in handcrafted features such as capitalization and external POS tags, and the decoder returns n-best outputs with probabilities under BIO or BIOES label schemes. That flexibility is why it still works as a research baseline, backed by an ACL 2018 demonstration paper and a best-paper report at COLING 2018, and it remains a fast way to prototype a labeling model on your own annotated data.